lustre/mdt/mdt_reint.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/mdt/mdt_reint.c
  32  *
  33  * Lustre Metadata Target (mdt) reintegration routines
  34  *
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  * Author: Huang Hua <huanghua@clusterfs.com>
  39  * Author: Yury Umanets <umka@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_MDS
  43
  44 #include <lprocfs_status.h>
  45 #include "mdt_internal.h"
  46 #include <lustre_lmv.h>
  47 #include <lustre_crypto.h>
  48
  49 static inline void mdt_reint_init_ma(struct mdt_thread_info *info,
  50                                      struct md_attr *ma)
  51 {
  52         ma->ma_need = MA_INODE;
  53         ma->ma_valid = 0;
  54 }
  55
  56 /**
  57  * Get version of object by fid.
  58  *
  59  * Return real version or ENOENT_VERSION if object doesn't exist
  60  */
  61 static void mdt_obj_version_get(struct mdt_thread_info *info,
  62                                 struct mdt_object *o, __u64 *version)
  63 {
  64         LASSERT(o);
  65
  66         if (mdt_object_exists(o) && !mdt_object_remote(o) &&
  67             !fid_is_obf(mdt_object_fid(o)))
  68                 *version = dt_version_get(info->mti_env, mdt_obj2dt(o));
  69         else
  70                 *version = ENOENT_VERSION;
  71         CDEBUG(D_INODE, "FID "DFID" version is %#llx\n",
  72                PFID(mdt_object_fid(o)), *version);
  73 }
  74
  75 /**
  76  * Check version is correct.
  77  *
  78  * Should be called only during replay.
  79  */
  80 static int mdt_version_check(struct ptlrpc_request *req,
  81                              __u64 version, int idx)
  82 {
  83         __u64 *pre_ver = lustre_msg_get_versions(req->rq_reqmsg);
  84
  85         ENTRY;
  86         if (!exp_connect_vbr(req->rq_export))
  87                 RETURN(0);
  88
  89         LASSERT(req_is_replay(req));
  90         /** VBR: version is checked always because costs nothing */
  91         LASSERT(idx < PTLRPC_NUM_VERSIONS);
  92         /** Sanity check for malformed buffers */
  93         if (pre_ver == NULL) {
  94                 CERROR("No versions in request buffer\n");
  95                 spin_lock(&req->rq_export->exp_lock);
  96                 req->rq_export->exp_vbr_failed = 1;
  97                 spin_unlock(&req->rq_export->exp_lock);
  98                 RETURN(-EOVERFLOW);
  99         } else if (pre_ver[idx] != version) {
 100                 CDEBUG(D_INODE, "Version mismatch %#llx != %#llx\n",
 101                        pre_ver[idx], version);
 102                 spin_lock(&req->rq_export->exp_lock);
 103                 req->rq_export->exp_vbr_failed = 1;
 104                 spin_unlock(&req->rq_export->exp_lock);
 105                 RETURN(-EOVERFLOW);
 106         }
 107         RETURN(0);
 108 }
 109
 110 /**
 111  * Save pre-versions in reply.
 112  */
 113 static void mdt_version_save(struct ptlrpc_request *req, __u64 version,
 114                              int idx)
 115 {
 116         __u64 *reply_ver;
 117
 118         if (!exp_connect_vbr(req->rq_export))
 119                 return;
 120
 121         LASSERT(!req_is_replay(req));
 122         LASSERT(req->rq_repmsg != NULL);
 123         reply_ver = lustre_msg_get_versions(req->rq_repmsg);
 124         if (reply_ver)
 125                 reply_ver[idx] = version;
 126 }
 127
 128 /**
 129  * Save enoent version, it is needed when it is obvious that object doesn't
 130  * exist, e.g. child during create.
 131  */
 132 static void mdt_enoent_version_save(struct mdt_thread_info *info, int idx)
 133 {
 134         /* save version of file name for replay, it must be ENOENT here */
 135         if (!req_is_replay(mdt_info_req(info))) {
 136                 info->mti_ver[idx] = ENOENT_VERSION;
 137                 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
 138         }
 139 }
 140
 141 /**
 142  * Get version from disk and save in reply buffer.
 143  *
 144  * Versions are saved in reply only during normal operations not replays.
 145  */
 146 void mdt_version_get_save(struct mdt_thread_info *info,
 147                           struct mdt_object *mto, int idx)
 148 {
 149         /* don't save versions during replay */
 150         if (!req_is_replay(mdt_info_req(info))) {
 151                 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
 152                 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
 153         }
 154 }
 155
 156 /**
 157  * Get version from disk and check it, no save in reply.
 158  */
 159 int mdt_version_get_check(struct mdt_thread_info *info,
 160                           struct mdt_object *mto, int idx)
 161 {
 162         /* only check versions during replay */
 163         if (!req_is_replay(mdt_info_req(info)))
 164                 return 0;
 165
 166         mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
 167         return mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
 168 }
 169
 170 /**
 171  * Get version from disk and check if recovery or just save.
 172  */
 173 int mdt_version_get_check_save(struct mdt_thread_info *info,
 174                                struct mdt_object *mto, int idx)
 175 {
 176         int rc = 0;
 177
 178         mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
 179         if (req_is_replay(mdt_info_req(info)))
 180                 rc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx],
 181                                        idx);
 182         else
 183                 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
 184         return rc;
 185 }
 186
 187 /**
 188  * Lookup with version checking.
 189  *
 190  * This checks version of 'name'. Many reint functions uses 'name' for child not
 191  * FID, therefore we need to get object by name and check its version.
 192  */
 193 int mdt_lookup_version_check(struct mdt_thread_info *info,
 194                              struct mdt_object *p,
 195                              const struct lu_name *lname,
 196                              struct lu_fid *fid, int idx)
 197 {
 198         int rc, vbrc;
 199
 200         rc = mdo_lookup(info->mti_env, mdt_object_child(p), lname, fid,
 201                         &info->mti_spec);
 202         /* Check version only during replay */
 203         if (!req_is_replay(mdt_info_req(info)))
 204                 return rc;
 205
 206         info->mti_ver[idx] = ENOENT_VERSION;
 207         if (rc == 0) {
 208                 struct mdt_object *child;
 209
 210                 child = mdt_object_find(info->mti_env, info->mti_mdt, fid);
 211                 if (likely(!IS_ERR(child))) {
 212                         mdt_obj_version_get(info, child, &info->mti_ver[idx]);
 213                         mdt_object_put(info->mti_env, child);
 214                 }
 215         }
 216         vbrc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
 217         return vbrc ? vbrc : rc;
 218
 219 }
 220
 221 static int mdt_unlock_slaves(struct mdt_thread_info *mti,
 222                              struct mdt_object *obj,
 223                              struct ldlm_enqueue_info *einfo,
 224                              int decref)
 225 {
 226         union ldlm_policy_data *policy = &mti->mti_policy;
 227         struct mdt_lock_handle *lh = &mti->mti_lh[MDT_LH_LOCAL];
 228         struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
 229         int i;
 230
 231         LASSERT(S_ISDIR(obj->mot_header.loh_attr));
 232         LASSERT(slave_locks);
 233
 234         memset(policy, 0, sizeof(*policy));
 235         policy->l_inodebits.bits = einfo->ei_inodebits;
 236         mdt_lock_handle_init(lh);
 237         mdt_lock_reg_init(lh, einfo->ei_mode);
 238         for (i = 0; i < slave_locks->ha_count; i++) {
 239                 if (test_bit(i, (void *)slave_locks->ha_map))
 240                         lh->mlh_rreg_lh = slave_locks->ha_handles[i];
 241                 else
 242                         lh->mlh_reg_lh = slave_locks->ha_handles[i];
 243                 mdt_object_unlock(mti, NULL, lh, decref);
 244                 slave_locks->ha_handles[i].cookie = 0ull;
 245         }
 246
 247         return mo_object_unlock(mti->mti_env, mdt_object_child(obj), einfo,
 248                                 policy);
 249 }
 250
 251 static inline int mdt_object_striped(struct mdt_thread_info *mti,
 252                                      struct mdt_object *obj)
 253 {
 254         struct lu_device *bottom_dev;
 255         struct lu_object *bottom_obj;
 256         int rc;
 257
 258         if (!S_ISDIR(obj->mot_header.loh_attr))
 259                 return 0;
 260
 261         /* getxattr from bottom obj to avoid reading in shard FIDs */
 262         bottom_dev = dt2lu_dev(mti->mti_mdt->mdt_bottom);
 263         bottom_obj = lu_object_find_slice(mti->mti_env, bottom_dev,
 264                                           mdt_object_fid(obj), NULL);
 265         if (IS_ERR(bottom_obj))
 266                 return PTR_ERR(bottom_obj);
 267
 268         rc = dt_xattr_get(mti->mti_env, lu2dt(bottom_obj), &LU_BUF_NULL,
 269                           XATTR_NAME_LMV);
 270         lu_object_put(mti->mti_env, bottom_obj);
 271
 272         return (rc > 0) ? 1 : (rc == -ENODATA) ? 0 : rc;
 273 }
 274
 275 /**
 276  * Lock slave stripes if necessary, the lock handles of slave stripes
 277  * will be stored in einfo->ei_cbdata.
 278  **/
 279 static int mdt_lock_slaves(struct mdt_thread_info *mti, struct mdt_object *obj,
 280                            enum ldlm_mode mode, __u64 ibits,
 281                            struct ldlm_enqueue_info *einfo)
 282 {
 283         union ldlm_policy_data *policy = &mti->mti_policy;
 284
 285         LASSERT(S_ISDIR(obj->mot_header.loh_attr));
 286
 287         einfo->ei_type = LDLM_IBITS;
 288         einfo->ei_mode = mode;
 289         einfo->ei_cb_bl = mdt_remote_blocking_ast;
 290         einfo->ei_cb_local_bl = mdt_blocking_ast;
 291         einfo->ei_cb_cp = ldlm_completion_ast;
 292         einfo->ei_enq_slave = 1;
 293         einfo->ei_namespace = mti->mti_mdt->mdt_namespace;
 294         einfo->ei_inodebits = ibits;
 295         einfo->ei_req_slot = 1;
 296         memset(policy, 0, sizeof(*policy));
 297         policy->l_inodebits.bits = ibits;
 298
 299         return mo_object_lock(mti->mti_env, mdt_object_child(obj), NULL, einfo,
 300                               policy);
 301 }
 302
 303 int mdt_reint_striped_lock(struct mdt_thread_info *info,
 304                            struct mdt_object *o,
 305                            struct mdt_lock_handle *lh,
 306                            __u64 ibits,
 307                            struct ldlm_enqueue_info *einfo,
 308                            bool cos_incompat)
 309 {
 310         int rc;
 311
 312         LASSERT(!mdt_object_remote(o));
 313
 314         memset(einfo, 0, sizeof(*einfo));
 315
 316         rc = mdt_reint_object_lock(info, o, lh, ibits, cos_incompat);
 317         if (rc)
 318                 return rc;
 319
 320         rc = mdt_object_striped(info, o);
 321         if (rc != 1) {
 322                 if (rc < 0)
 323                         mdt_object_unlock(info, o, lh, rc);
 324                 return rc;
 325         }
 326
 327         rc = mdt_lock_slaves(info, o, lh->mlh_reg_mode, ibits, einfo);
 328         if (rc) {
 329                 mdt_object_unlock(info, o, lh, rc);
 330                 if (rc == -EIO && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME))
 331                         rc = 0;
 332         }
 333
 334         return rc;
 335 }
 336
 337 void mdt_reint_striped_unlock(struct mdt_thread_info *info,
 338                               struct mdt_object *o,
 339                               struct mdt_lock_handle *lh,
 340                               struct ldlm_enqueue_info *einfo, int decref)
 341 {
 342         if (einfo->ei_cbdata)
 343                 mdt_unlock_slaves(info, o, einfo, decref);
 344         mdt_object_unlock(info, o, lh, decref);
 345 }
 346
 347 static int mdt_restripe(struct mdt_thread_info *info,
 348                         struct mdt_object *parent,
 349                         const struct lu_name *lname,
 350                         const struct lu_fid *tfid,
 351                         struct md_op_spec *spec,
 352                         struct md_attr *ma)
 353 {
 354         struct mdt_device *mdt = info->mti_mdt;
 355         struct lu_fid *fid = &info->mti_tmp_fid2;
 356         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
 357         struct lmv_user_md *lum = spec->u.sp_ea.eadata;
 358         struct lmv_mds_md_v1 *lmv;
 359         struct mdt_object *child;
 360         struct mdt_lock_handle *lhp;
 361         struct mdt_lock_handle *lhc;
 362         struct mdt_body *repbody;
 363         int rc;
 364
 365         ENTRY;
 366         if (!mdt->mdt_enable_dir_restripe)
 367                 RETURN(-EPERM);
 368
 369         LASSERT(lum);
 370         lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_FIXED);
 371
 372         rc = mdt_version_get_check_save(info, parent, 0);
 373         if (rc)
 374                 RETURN(rc);
 375
 376         lhp = &info->mti_lh[MDT_LH_PARENT];
 377         mdt_lock_pdo_init(lhp, LCK_PW, lname);
 378         rc = mdt_reint_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
 379                                    true);
 380         if (rc)
 381                 RETURN(rc);
 382
 383         rc = mdt_stripe_get(info, parent, ma, XATTR_NAME_LMV);
 384         if (rc)
 385                 GOTO(unlock_parent, rc);
 386
 387         if (ma->ma_valid & MA_LMV) {
 388                 /* don't allow restripe if parent dir layout is changing */
 389                 lmv = &ma->ma_lmv->lmv_md_v1;
 390                 if (!lmv_is_sane2(lmv))
 391                         GOTO(unlock_parent, rc = -EBADF);
 392
 393                 if (lmv_is_layout_changing(lmv))
 394                         GOTO(unlock_parent, rc = -EBUSY);
 395         }
 396
 397         fid_zero(fid);
 398         rc = mdt_lookup_version_check(info, parent, lname, fid, 1);
 399         if (rc)
 400                 GOTO(unlock_parent, rc);
 401
 402         child = mdt_object_find(info->mti_env, mdt, fid);
 403         if (IS_ERR(child))
 404                 GOTO(unlock_parent, rc = PTR_ERR(child));
 405
 406         if (!mdt_object_exists(child))
 407                 GOTO(out_child, rc = -ENOENT);
 408
 409         if (mdt_object_remote(child)) {
 410                 struct mdt_body *repbody;
 411
 412                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 413                 if (!repbody)
 414                         GOTO(out_child, rc = -EPROTO);
 415
 416                 repbody->mbo_fid1 = *fid;
 417                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
 418                 GOTO(out_child, rc = -EREMOTE);
 419         }
 420
 421         if (!S_ISDIR(lu_object_attr(&child->mot_obj)))
 422                 GOTO(out_child, rc = -ENOTDIR);
 423
 424         rc = mdt_stripe_get(info, child, ma, XATTR_NAME_LMV);
 425         if (rc)
 426                 GOTO(out_child, rc);
 427
 428         /* race with migrate? */
 429         if ((ma->ma_valid & MA_LMV) &&
 430              lmv_is_migrating(&ma->ma_lmv->lmv_md_v1))
 431                 GOTO(out_child, rc = -EBUSY);
 432
 433         /* lock object */
 434         lhc = &info->mti_lh[MDT_LH_CHILD];
 435         mdt_lock_reg_init(lhc, LCK_EX);
 436
 437         /* enqueue object remote LOOKUP lock */
 438         if (mdt_object_remote(parent)) {
 439                 rc = mdt_remote_object_lock(info, parent, fid,
 440                                             &lhc->mlh_rreg_lh,
 441                                             lhc->mlh_rreg_mode,
 442                                             MDS_INODELOCK_LOOKUP, false);
 443                 if (rc != ELDLM_OK)
 444                         GOTO(out_child, rc);
 445         }
 446
 447         rc = mdt_reint_striped_lock(info, child, lhc, MDS_INODELOCK_FULL, einfo,
 448                                     true);
 449         if (rc)
 450                 GOTO(unlock_child, rc);
 451
 452         tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
 453         rc = mdt_version_get_check_save(info, child, 1);
 454         if (rc)
 455                 GOTO(unlock_child, rc);
 456
 457         spin_lock(&mdt->mdt_restriper.mdr_lock);
 458         if (child->mot_restriping) {
 459                 /* race? */
 460                 spin_unlock(&mdt->mdt_restriper.mdr_lock);
 461                 GOTO(unlock_child, rc = -EBUSY);
 462         }
 463         child->mot_restriping = 1;
 464         spin_unlock(&mdt->mdt_restriper.mdr_lock);
 465
 466         *fid = *tfid;
 467         rc = mdt_restripe_internal(info, parent, child, lname, fid, spec, ma);
 468         if (rc)
 469                 GOTO(restriping_clear, rc);
 470
 471         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 472         if (!repbody)
 473                 GOTO(restriping_clear, rc = -EPROTO);
 474
 475         mdt_pack_attr2body(info, repbody, &ma->ma_attr, fid);
 476         EXIT;
 477
 478 restriping_clear:
 479         child->mot_restriping = 0;
 480 unlock_child:
 481         mdt_reint_striped_unlock(info, child, lhc, einfo, rc);
 482 out_child:
 483         mdt_object_put(info->mti_env, child);
 484 unlock_parent:
 485         mdt_object_unlock(info, parent, lhp, rc);
 486
 487         return rc;
 488 }
 489
 490 /*
 491  * VBR: we save three versions in reply:
 492  * 0 - parent. Check that parent version is the same during replay.
 493  * 1 - name. Version of 'name' if file exists with the same name or
 494  * ENOENT_VERSION, it is needed because file may appear due to missed replays.
 495  * 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
 496  * check.
 497  */
 498 static int mdt_create(struct mdt_thread_info *info)
 499 {
 500         struct mdt_device *mdt = info->mti_mdt;
 501         struct mdt_object *parent;
 502         struct mdt_object *child;
 503         struct mdt_lock_handle *lh;
 504         struct mdt_body *repbody;
 505         struct md_attr *ma = &info->mti_attr;
 506         struct mdt_reint_record *rr = &info->mti_rr;
 507         struct md_op_spec *spec = &info->mti_spec;
 508         bool restripe = false;
 509         int rc;
 510
 511         ENTRY;
 512         DEBUG_REQ(D_INODE, mdt_info_req(info),
 513                   "Create ("DNAME"->"DFID") in "DFID,
 514                   PNAME(&rr->rr_name), PFID(rr->rr_fid2), PFID(rr->rr_fid1));
 515
 516         if (!fid_is_md_operative(rr->rr_fid1))
 517                 RETURN(-EPERM);
 518
 519         if (S_ISDIR(ma->ma_attr.la_mode) &&
 520             spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
 521                 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
 522                 struct lu_ucred *uc = mdt_ucred(info);
 523                 struct obd_export *exp = mdt_info_req(info)->rq_export;
 524
 525                 /* Only new clients can create remote dir( >= 2.4) and
 526                  * striped dir(>= 2.6), old client will return -ENOTSUPP
 527                  */
 528                 if (!mdt_is_dne_client(exp))
 529                         RETURN(-ENOTSUPP);
 530
 531                 if (le32_to_cpu(lum->lum_stripe_count) > 1) {
 532                         if (!mdt_is_striped_client(exp))
 533                                 RETURN(-ENOTSUPP);
 534
 535                         if (!mdt->mdt_enable_striped_dir)
 536                                 RETURN(-EPERM);
 537                 } else if (!mdt->mdt_enable_remote_dir) {
 538                         RETURN(-EPERM);
 539                 }
 540
 541                 if ((!(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH)) &&
 542                     (le32_to_cpu(lum->lum_hash_type) & LMV_HASH_TYPE_MASK) ==
 543                     LMV_HASH_TYPE_CRUSH)
 544                         RETURN(-EPROTO);
 545
 546                 if (!cap_raised(uc->uc_cap, CAP_SYS_ADMIN) &&
 547                     uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
 548                     mdt->mdt_enable_remote_dir_gid != -1)
 549                         RETURN(-EPERM);
 550
 551                 /* restripe if later found dir exists, MDS_OPEN_CREAT means
 552                  * this is create only, don't try restripe.
 553                  */
 554                 if (mdt->mdt_enable_dir_restripe &&
 555                     le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT &&
 556                     !(spec->sp_cr_flags & MDS_OPEN_CREAT))
 557                         restripe = true;
 558         }
 559
 560         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 561
 562         parent = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
 563         if (IS_ERR(parent))
 564                 RETURN(PTR_ERR(parent));
 565
 566         if (!mdt_object_exists(parent))
 567                 GOTO(put_parent, rc = -ENOENT);
 568
 569         /*
 570          * LU-10235: check if name exists locklessly first to avoid massive
 571          * lock recalls on existing directories.
 572          */
 573         rc = mdt_lookup_version_check(info, parent, &rr->rr_name,
 574                                       &info->mti_tmp_fid1, 1);
 575         if (rc == 0) {
 576                 if (!restripe)
 577                         GOTO(put_parent, rc = -EEXIST);
 578
 579                 rc = mdt_restripe(info, parent, &rr->rr_name, rr->rr_fid2, spec,
 580                                   ma);
 581         }
 582
 583         /* -ENOENT is expected here */
 584         if (rc != -ENOENT)
 585                 GOTO(put_parent, rc);
 586
 587         /* save version of file name for replay, it must be ENOENT here */
 588         mdt_enoent_version_save(info, 1);
 589
 590         OBD_RACE(OBD_FAIL_MDS_CREATE_RACE);
 591
 592         lh = &info->mti_lh[MDT_LH_PARENT];
 593         mdt_lock_pdo_init(lh, LCK_PW, &rr->rr_name);
 594         rc = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
 595         if (rc)
 596                 GOTO(put_parent, rc);
 597
 598         if (!mdt_object_remote(parent)) {
 599                 rc = mdt_version_get_check_save(info, parent, 0);
 600                 if (rc)
 601                         GOTO(unlock_parent, rc);
 602         }
 603
 604         child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
 605         if (unlikely(IS_ERR(child)))
 606                 GOTO(unlock_parent, rc = PTR_ERR(child));
 607
 608         ma->ma_need = MA_INODE;
 609         ma->ma_valid = 0;
 610
 611         mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
 612                         OBD_FAIL_MDS_REINT_CREATE_WRITE);
 613
 614         /* Version of child will be updated on disk. */
 615         tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
 616         rc = mdt_version_get_check_save(info, child, 2);
 617         if (rc)
 618                 GOTO(put_child, rc);
 619
 620         /*
 621          * Do not perform lookup sanity check. We know that name does
 622          * not exist.
 623          */
 624         info->mti_spec.sp_cr_lookup = 0;
 625         info->mti_spec.sp_feat = &dt_directory_features;
 626
 627         rc = mdo_create(info->mti_env, mdt_object_child(parent), &rr->rr_name,
 628                         mdt_object_child(child), &info->mti_spec, ma);
 629         if (rc == 0)
 630                 rc = mdt_attr_get_complex(info, child, ma);
 631
 632         if (rc < 0)
 633                 GOTO(put_child, rc);
 634
 635         /*
 636          * On DNE, we need to eliminate dependey between 'mkdir a' and
 637          * 'mkdir a/b' if b is a striped directory, to achieve this, two
 638          * things are done below:
 639          * 1. save child and slaves lock.
 640          * 2. if the child is a striped directory, relock parent so to
 641          *    compare against with COS locks to ensure parent was
 642          *    committed to disk.
 643          */
 644         if (mdt_slc_is_enabled(mdt) && S_ISDIR(ma->ma_attr.la_mode)) {
 645                 struct mdt_lock_handle *lhc;
 646                 struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
 647                 bool cos_incompat;
 648
 649                 rc = mdt_object_striped(info, child);
 650                 if (rc < 0)
 651                         GOTO(put_child, rc);
 652
 653                 cos_incompat = rc;
 654                 if (cos_incompat) {
 655                         if (!mdt_object_remote(parent)) {
 656                                 mdt_object_unlock(info, parent, lh, 1);
 657                                 mdt_lock_pdo_init(lh, LCK_PW, &rr->rr_name);
 658                                 rc = mdt_reint_object_lock(info, parent, lh,
 659                                                            MDS_INODELOCK_UPDATE,
 660                                                            true);
 661                                 if (rc)
 662                                         GOTO(put_child, rc);
 663                         }
 664                 }
 665
 666                 lhc = &info->mti_lh[MDT_LH_CHILD];
 667                 mdt_lock_handle_init(lhc);
 668                 mdt_lock_reg_init(lhc, LCK_PW);
 669                 rc = mdt_reint_striped_lock(info, child, lhc,
 670                                             MDS_INODELOCK_UPDATE, einfo,
 671                                             cos_incompat);
 672                 if (rc)
 673                         GOTO(put_child, rc);
 674
 675                 mdt_reint_striped_unlock(info, child, lhc, einfo, rc);
 676         }
 677
 678         /* Return fid & attr to client. */
 679         if (ma->ma_valid & MA_INODE)
 680                 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
 681                                    mdt_object_fid(child));
 682         EXIT;
 683 put_child:
 684         mdt_object_put(info->mti_env, child);
 685 unlock_parent:
 686         mdt_object_unlock(info, parent, lh, rc);
 687 put_parent:
 688         mdt_object_put(info->mti_env, parent);
 689         return rc;
 690 }
 691
 692 static int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
 693                         struct md_attr *ma)
 694 {
 695         struct mdt_lock_handle  *lh;
 696         int do_vbr = ma->ma_attr.la_valid &
 697                         (LA_MODE | LA_UID | LA_GID | LA_PROJID | LA_FLAGS);
 698         __u64 lockpart = MDS_INODELOCK_UPDATE;
 699         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
 700         bool cos_incompat;
 701         int rc;
 702
 703         ENTRY;
 704         rc = mdt_object_striped(info, mo);
 705         if (rc < 0)
 706                 RETURN(rc);
 707
 708         cos_incompat = rc;
 709
 710         lh = &info->mti_lh[MDT_LH_PARENT];
 711         mdt_lock_reg_init(lh, LCK_PW);
 712
 713         /* Even though the new MDT will grant PERM lock to the old
 714          * client, but the old client will almost ignore that during
 715          * So it needs to revoke both LOOKUP and PERM lock here, so
 716          * both new and old client can cancel the dcache
 717          */
 718         if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
 719                 lockpart |= MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
 720
 721         rc = mdt_reint_striped_lock(info, mo, lh, lockpart, einfo,
 722                                     cos_incompat);
 723         if (rc != 0)
 724                 RETURN(rc);
 725
 726         /* all attrs are packed into mti_attr in unpack_setattr */
 727         mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
 728                        OBD_FAIL_MDS_REINT_SETATTR_WRITE);
 729
 730         /* VBR: update version if attr changed are important for recovery */
 731         if (do_vbr) {
 732                 /* update on-disk version of changed object */
 733                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mo));
 734                 rc = mdt_version_get_check_save(info, mo, 0);
 735                 if (rc)
 736                         GOTO(out_unlock, rc);
 737         }
 738
 739         /* Ensure constant striping during chown(). See LU-2789. */
 740         if (ma->ma_attr.la_valid & (LA_UID|LA_GID|LA_PROJID))
 741                 mutex_lock(&mo->mot_lov_mutex);
 742
 743         /* all attrs are packed into mti_attr in unpack_setattr */
 744         rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
 745
 746         if (ma->ma_attr.la_valid & (LA_UID|LA_GID|LA_PROJID))
 747                 mutex_unlock(&mo->mot_lov_mutex);
 748
 749         if (rc != 0)
 750                 GOTO(out_unlock, rc);
 751         mdt_dom_obj_lvb_update(info->mti_env, mo, NULL, false);
 752         EXIT;
 753 out_unlock:
 754         mdt_reint_striped_unlock(info, mo, lh, einfo, rc);
 755         return rc;
 756 }
 757
 758 /**
 759  * Check HSM flags and add HS_DIRTY flag if relevant.
 760  *
 761  * A file could be set dirty only if it has a copy in the backend (HS_EXISTS)
 762  * and is not RELEASED.
 763  */
 764 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
 765                         struct md_attr *ma)
 766 {
 767         struct lu_ucred *uc = mdt_ucred(info);
 768         kernel_cap_t cap_saved;
 769         int rc;
 770
 771         ENTRY;
 772         /* If the file was modified, add the dirty flag */
 773         ma->ma_need = MA_HSM;
 774         rc = mdt_attr_get_complex(info, mo, ma);
 775         if (rc) {
 776                 CERROR("file attribute read error for "DFID": %d.\n",
 777                         PFID(mdt_object_fid(mo)), rc);
 778                 RETURN(rc);
 779         }
 780
 781         /* If an up2date copy exists in the backend, add dirty flag */
 782         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_EXISTS)
 783             && !(ma->ma_hsm.mh_flags & (HS_DIRTY|HS_RELEASED))) {
 784                 ma->ma_hsm.mh_flags |= HS_DIRTY;
 785
 786                 /* Bump cap so that closes from non-owner writers can
 787                  * set the HSM state to dirty.
 788                  */
 789                 cap_saved = uc->uc_cap;
 790                 cap_raise(uc->uc_cap, CAP_FOWNER);
 791                 rc = mdt_hsm_attr_set(info, mo, &ma->ma_hsm);
 792                 uc->uc_cap = cap_saved;
 793                 if (rc)
 794                         CERROR("file attribute change error for "DFID": %d\n",
 795                                 PFID(mdt_object_fid(mo)), rc);
 796         }
 797
 798         RETURN(rc);
 799 }
 800
 801 static int mdt_reint_setattr(struct mdt_thread_info *info,
 802                              struct mdt_lock_handle *lhc)
 803 {
 804         struct mdt_device *mdt = info->mti_mdt;
 805         struct md_attr *ma = &info->mti_attr;
 806         struct mdt_reint_record *rr = &info->mti_rr;
 807         struct ptlrpc_request *req = mdt_info_req(info);
 808         struct mdt_object *mo;
 809         struct mdt_body *repbody;
 810         ktime_t kstart = ktime_get();
 811         int rc, rc2;
 812
 813         ENTRY;
 814         DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
 815                   (unsigned int)ma->ma_attr.la_valid);
 816
 817         if (info->mti_dlm_req)
 818                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
 819
 820         OBD_RACE(OBD_FAIL_PTLRPC_RESEND_RACE);
 821
 822         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 823         mo = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
 824         if (IS_ERR(mo))
 825                 GOTO(out, rc = PTR_ERR(mo));
 826
 827         if (!mdt_object_exists(mo))
 828                 GOTO(out_put, rc = -ENOENT);
 829
 830         if (mdt_object_remote(mo))
 831                 GOTO(out_put, rc = -EREMOTE);
 832
 833         ma->ma_enable_chprojid_gid = mdt->mdt_enable_chprojid_gid;
 834         /* revoke lease lock if size is going to be changed */
 835         if (unlikely(ma->ma_attr.la_valid & LA_SIZE &&
 836                      !(ma->ma_attr_flags & MDS_TRUNC_KEEP_LEASE) &&
 837                      atomic_read(&mo->mot_lease_count) > 0)) {
 838                 down_read(&mo->mot_open_sem);
 839
 840                 if (atomic_read(&mo->mot_lease_count) > 0) { /* lease exists */
 841                         lhc = &info->mti_lh[MDT_LH_LOCAL];
 842                         mdt_lock_reg_init(lhc, LCK_CW);
 843
 844                         rc = mdt_object_lock(info, mo, lhc, MDS_INODELOCK_OPEN);
 845                         if (rc != 0) {
 846                                 up_read(&mo->mot_open_sem);
 847                                 GOTO(out_put, rc);
 848                         }
 849
 850                         /* revoke lease lock */
 851                         mdt_object_unlock(info, mo, lhc, 1);
 852                 }
 853                 up_read(&mo->mot_open_sem);
 854         }
 855
 856         if (ma->ma_attr.la_valid & LA_SIZE || rr->rr_flags & MRF_OPEN_TRUNC) {
 857                 /* Check write access for the O_TRUNC case */
 858                 if (mdt_write_read(mo) < 0)
 859                         GOTO(out_put, rc = -ETXTBSY);
 860
 861                 /* LU-10286: compatibility check for FLR.
 862                  * Please check the comment in mdt_finish_open() for details
 863                  */
 864                 if (!exp_connect_flr(info->mti_exp) ||
 865                     !exp_connect_overstriping(info->mti_exp)) {
 866                         rc = mdt_big_xattr_get(info, mo, XATTR_NAME_LOV);
 867                         if (rc < 0 && rc != -ENODATA)
 868                                 GOTO(out_put, rc);
 869
 870                         if (!exp_connect_flr(info->mti_exp)) {
 871                                 if (rc > 0 &&
 872                                     mdt_lmm_is_flr(info->mti_big_lmm))
 873                                         GOTO(out_put, rc = -EOPNOTSUPP);
 874                         }
 875
 876                         if (!exp_connect_overstriping(info->mti_exp)) {
 877                                 if (rc > 0 &&
 878                                     mdt_lmm_is_overstriping(info->mti_big_lmm))
 879                                         GOTO(out_put, rc = -EOPNOTSUPP);
 880                         }
 881                 }
 882
 883                 /* For truncate, the file size sent from client
 884                  * is believable, but the blocks are incorrect,
 885                  * which makes the block size in LSOM attribute
 886                  * inconsisent with the real block size.
 887                  */
 888                 rc = mdt_lsom_update(info, mo, true);
 889                 if (rc)
 890                         GOTO(out_put, rc);
 891         }
 892
 893         if ((ma->ma_valid & MA_INODE) && ma->ma_attr.la_valid) {
 894                 if (ma->ma_valid & MA_LOV)
 895                         GOTO(out_put, rc = -EPROTO);
 896
 897                 /* MDT supports FMD for regular files due to Data-on-MDT */
 898                 if (S_ISREG(lu_object_attr(&mo->mot_obj)) &&
 899                     ma->ma_attr.la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
 900                         tgt_fmd_update(info->mti_exp, mdt_object_fid(mo),
 901                                        req->rq_xid);
 902
 903                         if (ma->ma_attr.la_valid & LA_MTIME) {
 904                                 rc = mdt_attr_get_pfid(info, mo, &ma->ma_pfid);
 905                                 if (!rc)
 906                                         ma->ma_valid |= MA_PFID;
 907                         }
 908                 }
 909
 910                 rc = mdt_attr_set(info, mo, ma);
 911                 if (rc)
 912                         GOTO(out_put, rc);
 913         } else if ((ma->ma_valid & (MA_LOV | MA_LMV)) &&
 914                    (ma->ma_valid & MA_INODE)) {
 915                 struct lu_buf *buf = &info->mti_buf;
 916                 struct lu_ucred *uc = mdt_ucred(info);
 917                 struct mdt_lock_handle *lh;
 918                 const char *name;
 919                 __u64 lockpart = MDS_INODELOCK_XATTR;
 920
 921                 /* reject if either remote or striped dir is disabled */
 922                 if (ma->ma_valid & MA_LMV) {
 923                         if (!mdt->mdt_enable_remote_dir ||
 924                             !mdt->mdt_enable_striped_dir)
 925                                 GOTO(out_put, rc = -EPERM);
 926
 927                         if (!cap_raised(uc->uc_cap, CAP_SYS_ADMIN) &&
 928                             uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
 929                             mdt->mdt_enable_remote_dir_gid != -1)
 930                                 GOTO(out_put, rc = -EPERM);
 931                 }
 932
 933                 if (!S_ISDIR(lu_object_attr(&mo->mot_obj)))
 934                         GOTO(out_put, rc = -ENOTDIR);
 935
 936                 if (ma->ma_attr.la_valid != 0)
 937                         GOTO(out_put, rc = -EPROTO);
 938
 939                 lh = &info->mti_lh[MDT_LH_PARENT];
 940                 mdt_lock_reg_init(lh, LCK_PW);
 941
 942                 if (ma->ma_valid & MA_LOV) {
 943                         buf->lb_buf = ma->ma_lmm;
 944                         buf->lb_len = ma->ma_lmm_size;
 945                         name = XATTR_NAME_LOV;
 946                 } else {
 947                         struct lmv_user_md *lmu = &ma->ma_lmv->lmv_user_md;
 948                         struct lu_fid *pfid = &info->mti_tmp_fid1;
 949                         struct lu_name *pname = &info->mti_name;
 950                         const char dotdot[] = "..";
 951                         struct mdt_object *pobj;
 952
 953                         buf->lb_buf = lmu;
 954                         buf->lb_len = ma->ma_lmv_size;
 955                         name = XATTR_NAME_DEFAULT_LMV;
 956
 957                         if (fid_is_root(rr->rr_fid1)) {
 958                                 lockpart |= MDS_INODELOCK_LOOKUP;
 959                         } else {
 960                                 /* force client to update dir default layout */
 961                                 fid_zero(pfid);
 962                                 pname->ln_name = dotdot;
 963                                 pname->ln_namelen = sizeof(dotdot);
 964                                 rc = mdo_lookup(info->mti_env,
 965                                                 mdt_object_child(mo), pname,
 966                                                 pfid, NULL);
 967                                 if (rc)
 968                                         GOTO(out_put, rc);
 969
 970                                 pobj = mdt_object_find(info->mti_env, mdt,
 971                                                        pfid);
 972                                 if (IS_ERR(pobj))
 973                                         GOTO(out_put, rc = PTR_ERR(pobj));
 974
 975                                 if (mdt_object_remote(pobj))
 976                                         rc = mdt_remote_object_lock(info, pobj,
 977                                                 mdt_object_fid(mo),
 978                                                 &lh->mlh_rreg_lh, LCK_EX,
 979                                                 MDS_INODELOCK_LOOKUP, false);
 980                                 else
 981                                         lockpart |= MDS_INODELOCK_LOOKUP;
 982
 983                                 mdt_object_put(info->mti_env, pobj);
 984
 985                                 if (rc)
 986                                         GOTO(out_put, rc);
 987                         }
 988                 }
 989
 990                 rc = mdt_object_lock(info, mo, lh, lockpart);
 991                 if (rc != 0)
 992                         GOTO(out_put, rc);
 993
 994                 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo), buf,
 995                                   name, 0);
 996
 997                 mdt_object_unlock(info, mo, lh, rc);
 998                 if (rc)
 999                         GOTO(out_put, rc);
1000         } else {
1001                 GOTO(out_put, rc = -EPROTO);
1002         }
1003
1004         /* If file data is modified, add the dirty flag */
1005         if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
1006                 rc = mdt_add_dirty_flag(info, mo, ma);
1007
1008         ma->ma_need = MA_INODE;
1009         ma->ma_valid = 0;
1010         rc = mdt_attr_get_complex(info, mo, ma);
1011         if (rc != 0)
1012                 GOTO(out_put, rc);
1013
1014         mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
1015
1016         EXIT;
1017 out_put:
1018         mdt_object_put(info->mti_env, mo);
1019 out:
1020         if (rc == 0)
1021                 mdt_counter_incr(req, LPROC_MDT_SETATTR,
1022                                  ktime_us_delta(ktime_get(), kstart));
1023
1024         mdt_client_compatibility(info);
1025         rc2 = mdt_fix_reply(info);
1026         if (rc == 0)
1027                 rc = rc2;
1028         return rc;
1029 }
1030
1031 static int mdt_reint_create(struct mdt_thread_info *info,
1032                             struct mdt_lock_handle *lhc)
1033 {
1034         struct ptlrpc_request   *req = mdt_info_req(info);
1035         ktime_t                 kstart = ktime_get();
1036         int                     rc;
1037
1038         ENTRY;
1039         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
1040                 RETURN(err_serious(-ESTALE));
1041
1042         if (info->mti_dlm_req)
1043                 ldlm_request_cancel(mdt_info_req(info),
1044                                     info->mti_dlm_req, 0, LATF_SKIP);
1045
1046         if (!lu_name_is_valid(&info->mti_rr.rr_name))
1047                 RETURN(-EPROTO);
1048
1049         switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
1050         case S_IFDIR:
1051         case S_IFREG:
1052         case S_IFLNK:
1053         case S_IFCHR:
1054         case S_IFBLK:
1055         case S_IFIFO:
1056         case S_IFSOCK:
1057                 break;
1058         default:
1059                 CERROR("%s: Unsupported mode %o\n",
1060                        mdt_obd_name(info->mti_mdt),
1061                        info->mti_attr.ma_attr.la_mode);
1062                 RETURN(err_serious(-EOPNOTSUPP));
1063         }
1064
1065         rc = mdt_create(info);
1066         if (rc == 0) {
1067                 if ((info->mti_attr.ma_attr.la_mode & S_IFMT) == S_IFDIR)
1068                         mdt_counter_incr(req, LPROC_MDT_MKDIR,
1069                                          ktime_us_delta(ktime_get(), kstart));
1070                 else
1071                         /* Special file should stay on the same node as parent*/
1072                         mdt_counter_incr(req, LPROC_MDT_MKNOD,
1073                                          ktime_us_delta(ktime_get(), kstart));
1074         }
1075
1076         RETURN(rc);
1077 }
1078
1079 /*
1080  * VBR: save parent version in reply and child version getting by its name.
1081  * Version of child is getting and checking during its lookup. If
1082  */
1083 static int mdt_reint_unlink(struct mdt_thread_info *info,
1084                             struct mdt_lock_handle *lhc)
1085 {
1086         struct mdt_reint_record *rr = &info->mti_rr;
1087         struct ptlrpc_request *req = mdt_info_req(info);
1088         struct md_attr *ma = &info->mti_attr;
1089         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1090         struct mdt_object *mp;
1091         struct mdt_object *mc;
1092         struct mdt_lock_handle *parent_lh;
1093         struct mdt_lock_handle *child_lh;
1094         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
1095         __u64 lock_ibits;
1096         bool cos_incompat = false;
1097         int no_name = 0;
1098         ktime_t kstart = ktime_get();
1099         int rc;
1100
1101         ENTRY;
1102         DEBUG_REQ(D_INODE, req, "unlink "DFID"/"DNAME"", PFID(rr->rr_fid1),
1103                   PNAME(&rr->rr_name));
1104
1105         if (info->mti_dlm_req)
1106                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1107
1108         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
1109                 RETURN(err_serious(-ENOENT));
1110
1111         if (!fid_is_md_operative(rr->rr_fid1))
1112                 RETURN(-EPERM);
1113
1114         mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1115         if (IS_ERR(mp))
1116                 RETURN(PTR_ERR(mp));
1117
1118         if (mdt_object_remote(mp)) {
1119                 cos_incompat = true;
1120         } else {
1121                 rc = mdt_version_get_check_save(info, mp, 0);
1122                 if (rc)
1123                         GOTO(put_parent, rc);
1124         }
1125
1126         OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
1127         OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
1128 relock:
1129         parent_lh = &info->mti_lh[MDT_LH_PARENT];
1130         mdt_lock_pdo_init(parent_lh, LCK_PW, &rr->rr_name);
1131         rc = mdt_reint_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
1132                                    cos_incompat);
1133         if (rc != 0)
1134                 GOTO(put_parent, rc);
1135
1136         if (info->mti_spec.sp_cr_flags & MDS_OP_WITH_FID) {
1137                 *child_fid = *rr->rr_fid2;
1138         } else {
1139                 /* lookup child object along with version checking */
1140                 fid_zero(child_fid);
1141                 rc = mdt_lookup_version_check(info, mp, &rr->rr_name, child_fid,
1142                                               1);
1143                 if (rc != 0) {
1144                         /* Name might not be able to find during resend of
1145                          * remote unlink, considering following case.
1146                          * dir_A is a remote directory, the name entry of
1147                          * dir_A is on MDT0, the directory is on MDT1,
1148                          *
1149                          * 1. client sends unlink req to MDT1.
1150                          * 2. MDT1 sends name delete update to MDT0.
1151                          * 3. name entry is being deleted in MDT0 synchronously.
1152                          * 4. MDT1 is restarted.
1153                          * 5. client resends unlink req to MDT1. So it can not
1154                          *    find the name entry on MDT0 anymore.
1155                          * In this case, MDT1 only needs to destory the local
1156                          * directory.
1157                          */
1158                         if (mdt_object_remote(mp) && rc == -ENOENT &&
1159                             !fid_is_zero(rr->rr_fid2) &&
1160                             lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1161                                 no_name = 1;
1162                                 *child_fid = *rr->rr_fid2;
1163                         } else {
1164                                 GOTO(unlock_parent, rc);
1165                         }
1166                 }
1167         }
1168
1169         if (!fid_is_md_operative(child_fid))
1170                 GOTO(unlock_parent, rc = -EPERM);
1171
1172         /* We will lock the child regardless it is local or remote. No harm. */
1173         mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1174         if (IS_ERR(mc))
1175                 GOTO(unlock_parent, rc = PTR_ERR(mc));
1176
1177         if (info->mti_spec.sp_cr_flags & MDS_OP_WITH_FID) {
1178                 /* In this case, child fid is embedded in the request, and we do
1179                  * not have a proper name as rr_name contains an encoded
1180                  * hash. So find name that matches provided hash.
1181                  */
1182                 if (!find_name_matching_hash(info, &rr->rr_name,
1183                                              NULL, mc, false))
1184                         GOTO(put_child, rc = -ENOENT);
1185         }
1186
1187         if (!cos_incompat) {
1188                 rc = mdt_object_striped(info, mc);
1189                 if (rc < 0)
1190                         GOTO(put_child, rc);
1191
1192                 cos_incompat = rc;
1193                 if (cos_incompat) {
1194                         mdt_object_put(info->mti_env, mc);
1195                         mdt_object_unlock(info, mp, parent_lh, -EAGAIN);
1196                         goto relock;
1197                 }
1198         }
1199
1200         child_lh = &info->mti_lh[MDT_LH_CHILD];
1201         mdt_lock_reg_init(child_lh, LCK_EX);
1202         if (info->mti_spec.sp_rm_entry) {
1203                 struct lu_ucred *uc  = mdt_ucred(info);
1204
1205                 if (!mdt_is_dne_client(req->rq_export))
1206                         /* Return -ENOTSUPP for old client */
1207                         GOTO(put_child, rc = -ENOTSUPP);
1208
1209                 if (!cap_raised(uc->uc_cap, CAP_SYS_ADMIN))
1210                         GOTO(put_child, rc = -EPERM);
1211
1212                 ma->ma_need = MA_INODE;
1213                 ma->ma_valid = 0;
1214                 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1215                                 NULL, &rr->rr_name, ma, no_name);
1216                 GOTO(put_child, rc);
1217         }
1218
1219         if (mdt_object_remote(mc)) {
1220                 struct mdt_body  *repbody;
1221
1222                 if (!fid_is_zero(rr->rr_fid2)) {
1223                         CDEBUG(D_INFO, "%s: name "DNAME" cannot find "DFID"\n",
1224                                mdt_obd_name(info->mti_mdt),
1225                                PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
1226                         GOTO(put_child, rc = -ENOENT);
1227                 }
1228                 CDEBUG(D_INFO, "%s: name "DNAME": "DFID" is on another MDT\n",
1229                        mdt_obd_name(info->mti_mdt),
1230                        PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
1231
1232                 if (!mdt_is_dne_client(req->rq_export))
1233                         /* Return -ENOTSUPP for old client */
1234                         GOTO(put_child, rc = -ENOTSUPP);
1235
1236                 /* Revoke the LOOKUP lock of the remote object granted by
1237                  * this MDT. Since the unlink will happen on another MDT,
1238                  * it will release the LOOKUP lock right away. Then What
1239                  * would happen if another client try to grab the LOOKUP
1240                  * lock at the same time with unlink XXX
1241                  */
1242                 mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP);
1243                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1244                 LASSERT(repbody != NULL);
1245                 repbody->mbo_fid1 = *mdt_object_fid(mc);
1246                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1247                 GOTO(unlock_child, rc = -EREMOTE);
1248         }
1249         /* We used to acquire MDS_INODELOCK_FULL here but we can't do
1250          * this now because a running HSM restore on the child (unlink
1251          * victim) will hold the layout lock. See LU-4002.
1252          */
1253         lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
1254         if (mdt_object_remote(mp)) {
1255                 /* Enqueue lookup lock from parent MDT */
1256                 rc = mdt_remote_object_lock(info, mp, mdt_object_fid(mc),
1257                                             &child_lh->mlh_rreg_lh,
1258                                             child_lh->mlh_rreg_mode,
1259                                             MDS_INODELOCK_LOOKUP, false);
1260                 if (rc != ELDLM_OK)
1261                         GOTO(put_child, rc);
1262
1263                 lock_ibits &= ~MDS_INODELOCK_LOOKUP;
1264         }
1265
1266         rc = mdt_reint_striped_lock(info, mc, child_lh, lock_ibits, einfo,
1267                                     cos_incompat);
1268         if (rc != 0)
1269                 GOTO(put_child, rc);
1270
1271         /*
1272          * Now we can only make sure we need MA_INODE, in mdd layer, will check
1273          * whether need MA_LOV and MA_COOKIE.
1274          */
1275         ma->ma_need = MA_INODE;
1276         ma->ma_valid = 0;
1277
1278         mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1279                        OBD_FAIL_MDS_REINT_UNLINK_WRITE);
1280         /* save version when object is locked */
1281         mdt_version_get_save(info, mc, 1);
1282
1283         mutex_lock(&mc->mot_lov_mutex);
1284
1285         rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1286                         mdt_object_child(mc), &rr->rr_name, ma, no_name);
1287
1288         mutex_unlock(&mc->mot_lov_mutex);
1289         if (rc != 0)
1290                 GOTO(unlock_child, rc);
1291
1292         if (!lu_object_is_dying(&mc->mot_header)) {
1293                 rc = mdt_attr_get_complex(info, mc, ma);
1294                 if (rc)
1295                         GOTO(out_stat, rc);
1296         } else if (mdt_dom_check_for_discard(info, mc)) {
1297                 mdt_dom_discard_data(info, mc);
1298         }
1299         mdt_handle_last_unlink(info, mc, ma);
1300
1301 out_stat:
1302         if (ma->ma_valid & MA_INODE) {
1303                 switch (ma->ma_attr.la_mode & S_IFMT) {
1304                 case S_IFDIR:
1305                         mdt_counter_incr(req, LPROC_MDT_RMDIR,
1306                                          ktime_us_delta(ktime_get(), kstart));
1307                         break;
1308                 case S_IFREG:
1309                 case S_IFLNK:
1310                 case S_IFCHR:
1311                 case S_IFBLK:
1312                 case S_IFIFO:
1313                 case S_IFSOCK:
1314                         mdt_counter_incr(req, LPROC_MDT_UNLINK,
1315                                          ktime_us_delta(ktime_get(), kstart));
1316                         break;
1317                 default:
1318                         LASSERTF(0, "bad file type %o unlinking\n",
1319                                 ma->ma_attr.la_mode);
1320                 }
1321         }
1322
1323         EXIT;
1324
1325 unlock_child:
1326         mdt_reint_striped_unlock(info, mc, child_lh, einfo, rc);
1327 put_child:
1328         if (info->mti_spec.sp_cr_flags & MDS_OP_WITH_FID &&
1329             info->mti_big_buf.lb_buf)
1330                 lu_buf_free(&info->mti_big_buf);
1331         mdt_object_put(info->mti_env, mc);
1332 unlock_parent:
1333         mdt_object_unlock(info, mp, parent_lh, rc);
1334 put_parent:
1335         mdt_object_put(info->mti_env, mp);
1336         CFS_RACE_WAKEUP(OBD_FAIL_OBD_ZERO_NLINK_RACE);
1337         return rc;
1338 }
1339
1340 /*
1341  * VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
1342  * name.
1343  */
1344 static int mdt_reint_link(struct mdt_thread_info *info,
1345                           struct mdt_lock_handle *lhc)
1346 {
1347         struct mdt_reint_record *rr = &info->mti_rr;
1348         struct ptlrpc_request   *req = mdt_info_req(info);
1349         struct md_attr          *ma = &info->mti_attr;
1350         struct mdt_object       *ms;
1351         struct mdt_object       *mp;
1352         struct mdt_lock_handle  *lhs;
1353         struct mdt_lock_handle  *lhp;
1354         ktime_t kstart = ktime_get();
1355         bool cos_incompat;
1356         int rc;
1357
1358         ENTRY;
1359         DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/"DNAME,
1360                   PFID(rr->rr_fid1), PFID(rr->rr_fid2), PNAME(&rr->rr_name));
1361
1362         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
1363                 RETURN(err_serious(-ENOENT));
1364
1365         if (OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_RESEND_RACE) ||
1366             OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND)) {
1367                 req->rq_no_reply = 1;
1368                 RETURN(err_serious(-ENOENT));
1369         }
1370
1371         if (info->mti_dlm_req)
1372                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1373
1374         /* Invalid case so return error immediately instead of
1375          * processing it
1376          */
1377         if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
1378                 RETURN(-EPERM);
1379
1380         if (!fid_is_md_operative(rr->rr_fid1) ||
1381             !fid_is_md_operative(rr->rr_fid2))
1382                 RETURN(-EPERM);
1383
1384         /* step 1: find target parent dir */
1385         mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
1386         if (IS_ERR(mp))
1387                 RETURN(PTR_ERR(mp));
1388
1389         rc = mdt_version_get_check_save(info, mp, 0);
1390         if (rc)
1391                 GOTO(put_parent, rc);
1392
1393         /* step 2: find source */
1394         ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1395         if (IS_ERR(ms))
1396                 GOTO(put_parent, rc = PTR_ERR(ms));
1397
1398         if (!mdt_object_exists(ms)) {
1399                 CDEBUG(D_INFO, "%s: "DFID" does not exist.\n",
1400                        mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1));
1401                 GOTO(put_source, rc = -ENOENT);
1402         }
1403
1404         cos_incompat = (mdt_object_remote(mp) || mdt_object_remote(ms));
1405
1406         OBD_RACE(OBD_FAIL_MDS_LINK_RENAME_RACE);
1407
1408         lhp = &info->mti_lh[MDT_LH_PARENT];
1409         mdt_lock_pdo_init(lhp, LCK_PW, &rr->rr_name);
1410         rc = mdt_reint_object_lock(info, mp, lhp, MDS_INODELOCK_UPDATE,
1411                                    cos_incompat);
1412         if (rc != 0)
1413                 GOTO(put_source, rc);
1414
1415         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1416
1417         lhs = &info->mti_lh[MDT_LH_CHILD];
1418         mdt_lock_reg_init(lhs, LCK_EX);
1419         rc = mdt_reint_object_lock(info, ms, lhs,
1420                                    MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1421                                    cos_incompat);
1422         if (rc != 0)
1423                 GOTO(unlock_parent, rc);
1424
1425         /* step 3: link it */
1426         mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1427                         OBD_FAIL_MDS_REINT_LINK_WRITE);
1428
1429         tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
1430         rc = mdt_version_get_check_save(info, ms, 1);
1431         if (rc)
1432                 GOTO(unlock_source, rc);
1433
1434         /** check target version by name during replay */
1435         rc = mdt_lookup_version_check(info, mp, &rr->rr_name,
1436                                       &info->mti_tmp_fid1, 2);
1437         if (rc != 0 && rc != -ENOENT)
1438                 GOTO(unlock_source, rc);
1439         /* save version of file name for replay, it must be ENOENT here */
1440         if (!req_is_replay(mdt_info_req(info))) {
1441                 if (rc != -ENOENT) {
1442                         CDEBUG(D_INFO, "link target "DNAME" existed!\n",
1443                                PNAME(&rr->rr_name));
1444                         GOTO(unlock_source, rc = -EEXIST);
1445                 }
1446                 info->mti_ver[2] = ENOENT_VERSION;
1447                 mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
1448         }
1449
1450         rc = mdo_link(info->mti_env, mdt_object_child(mp),
1451                       mdt_object_child(ms), &rr->rr_name, ma);
1452
1453         if (rc == 0)
1454                 mdt_counter_incr(req, LPROC_MDT_LINK,
1455                                  ktime_us_delta(ktime_get(), kstart));
1456
1457         EXIT;
1458 unlock_source:
1459         mdt_object_unlock(info, ms, lhs, rc);
1460 unlock_parent:
1461         mdt_object_unlock(info, mp, lhp, rc);
1462 put_source:
1463         mdt_object_put(info->mti_env, ms);
1464 put_parent:
1465         mdt_object_put(info->mti_env, mp);
1466         return rc;
1467 }
1468 /**
1469  * lock the part of the directory according to the hash of the name
1470  * (lh->mlh_pdo_hash) in parallel directory lock.
1471  */
1472 static int mdt_pdir_hash_lock(struct mdt_thread_info *info,
1473                               struct mdt_lock_handle *lh,
1474                               struct mdt_object *obj, __u64 ibits,
1475                               bool cos_incompat)
1476 {
1477         struct ldlm_res_id *res = &info->mti_res_id;
1478         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1479         union ldlm_policy_data *policy = &info->mti_policy;
1480         __u64 dlmflags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1481         int rc;
1482
1483         /*
1484          * Finish res_id initializing by name hash marking part of
1485          * directory which is taking modification.
1486          */
1487         LASSERT(lh->mlh_pdo_hash != 0);
1488         fid_build_pdo_res_name(mdt_object_fid(obj), lh->mlh_pdo_hash, res);
1489         memset(policy, 0, sizeof(*policy));
1490         policy->l_inodebits.bits = ibits;
1491         if (cos_incompat &&
1492             (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
1493                 dlmflags |= LDLM_FL_COS_INCOMPAT;
1494         /*
1495          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1496          * going to be sent to client. If it is - mdt_intent_policy() path will
1497          * fix it up and turn FL_LOCAL flag off.
1498          */
1499         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
1500                           policy, res, dlmflags,
1501                           &info->mti_exp->exp_handle.h_cookie);
1502         return rc;
1503 }
1504
1505 /**
1506  * Get BFL lock for rename or migrate process.
1507  **/
1508 static int mdt_rename_lock(struct mdt_thread_info *info,
1509                            struct lustre_handle *lh)
1510 {
1511         int     rc;
1512
1513         ENTRY;
1514         if (mdt_seq_site(info->mti_mdt)->ss_node_id != 0) {
1515                 struct lu_fid *fid = &info->mti_tmp_fid1;
1516                 struct mdt_object *obj;
1517
1518                 /* XXX, right now, it has to use object API to
1519                  * enqueue lock cross MDT, so it will enqueue
1520                  * rename lock(with LUSTRE_BFL_FID) by root object
1521                  */
1522                 lu_root_fid(fid);
1523                 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1524                 if (IS_ERR(obj))
1525                         RETURN(PTR_ERR(obj));
1526
1527                 rc = mdt_remote_object_lock(info, obj,
1528                                             &LUSTRE_BFL_FID, lh,
1529                                             LCK_EX,
1530                                             MDS_INODELOCK_UPDATE, false);
1531                 mdt_object_put(info->mti_env, obj);
1532         } else {
1533                 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1534                 union ldlm_policy_data *policy = &info->mti_policy;
1535                 struct ldlm_res_id *res_id = &info->mti_res_id;
1536                 __u64 flags = 0;
1537
1538                 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
1539                 memset(policy, 0, sizeof(*policy));
1540                 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1541                 flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1542                 rc = ldlm_cli_enqueue_local(info->mti_env, ns, res_id,
1543                                             LDLM_IBITS, policy, LCK_EX, &flags,
1544                                             ldlm_blocking_ast,
1545                                             ldlm_completion_ast, NULL, NULL, 0,
1546                                             LVB_T_NONE,
1547                                             &info->mti_exp->exp_handle.h_cookie,
1548                                             lh);
1549                 RETURN(rc);
1550         }
1551         RETURN(rc);
1552 }
1553
1554 static void mdt_rename_unlock(struct lustre_handle *lh)
1555 {
1556         ENTRY;
1557         LASSERT(lustre_handle_is_used(lh));
1558         /* Cancel the single rename lock right away */
1559         ldlm_lock_decref_and_cancel(lh, LCK_EX);
1560         EXIT;
1561 }
1562
1563 static struct mdt_object *mdt_parent_find_check(struct mdt_thread_info *info,
1564                                                 const struct lu_fid *fid,
1565                                                 int idx)
1566 {
1567         struct mdt_object *dir;
1568         int rc;
1569
1570         ENTRY;
1571         dir = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1572         if (IS_ERR(dir))
1573                 RETURN(dir);
1574
1575         /* check early, the real version will be saved after locking */
1576         rc = mdt_version_get_check(info, dir, idx);
1577         if (rc)
1578                 GOTO(out_put, rc);
1579
1580         if (!mdt_object_exists(dir))
1581                 GOTO(out_put, rc = -ENOENT);
1582
1583         if (!S_ISDIR(lu_object_attr(&dir->mot_obj)))
1584                 GOTO(out_put, rc = -ENOTDIR);
1585
1586         RETURN(dir);
1587 out_put:
1588         mdt_object_put(info->mti_env, dir);
1589         return ERR_PTR(rc);
1590 }
1591
1592 /*
1593  * in case obj is remote obj on its parent, revoke LOOKUP lock,
1594  * herein we don't really check it, just do revoke.
1595  */
1596 int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
1597                                   struct mdt_object *pobj,
1598                                   struct mdt_object *obj)
1599 {
1600         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1601         int rc;
1602
1603         mdt_lock_handle_init(lh);
1604         mdt_lock_reg_init(lh, LCK_EX);
1605
1606         if (mdt_object_remote(pobj)) {
1607                 /* don't bother to check if pobj and obj are on the same MDT. */
1608                 rc = mdt_remote_object_lock(info, pobj, mdt_object_fid(obj),
1609                                             &lh->mlh_rreg_lh, LCK_EX,
1610                                             MDS_INODELOCK_LOOKUP, false);
1611         } else if (mdt_object_remote(obj)) {
1612                 struct ldlm_res_id *res = &info->mti_res_id;
1613                 union ldlm_policy_data *policy = &info->mti_policy;
1614                 __u64 dlmflags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB |
1615                                  LDLM_FL_COS_INCOMPAT;
1616
1617                 fid_build_reg_res_name(mdt_object_fid(obj), res);
1618                 memset(policy, 0, sizeof(*policy));
1619                 policy->l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1620                 rc = mdt_fid_lock(info->mti_env, info->mti_mdt->mdt_namespace,
1621                                   &lh->mlh_reg_lh, LCK_EX, policy, res,
1622                                   dlmflags, NULL);
1623         } else {
1624                 /* do nothing if both are local */
1625                 return 0;
1626         }
1627
1628         if (rc != ELDLM_OK)
1629                 return rc;
1630
1631         /*
1632          * TODO, currently we don't save this lock because there is no place to
1633          * hold this lock handle, but to avoid race we need to save this lock.
1634          */
1635         mdt_object_unlock(info, NULL, lh, 1);
1636
1637         return 0;
1638 }
1639
1640 /*
1641  * operation may takes locks of linkea, or directory stripes, group them in
1642  * different list.
1643  */
1644 struct mdt_sub_lock {
1645         struct mdt_object *msl_obj;
1646         struct mdt_lock_handle msl_lh;
1647         struct list_head msl_linkage;
1648 };
1649
1650 static void mdt_unlock_list(struct mdt_thread_info *info,
1651                             struct list_head *list, int decref)
1652 {
1653         struct mdt_sub_lock *msl;
1654         struct mdt_sub_lock *tmp;
1655
1656         list_for_each_entry_safe(msl, tmp, list, msl_linkage) {
1657                 mdt_object_unlock_put(info, msl->msl_obj, &msl->msl_lh, decref);
1658                 list_del(&msl->msl_linkage);
1659                 OBD_FREE_PTR(msl);
1660         }
1661 }
1662
1663 static inline void mdt_migrate_object_unlock(struct mdt_thread_info *info,
1664                                              struct mdt_object *obj,
1665                                              struct mdt_lock_handle *lh,
1666                                              struct ldlm_enqueue_info *einfo,
1667                                              struct list_head *slave_locks,
1668                                              int decref)
1669 {
1670         if (mdt_object_remote(obj)) {
1671                 mdt_unlock_list(info, slave_locks, decref);
1672                 mdt_object_unlock(info, obj, lh, decref);
1673         } else {
1674                 mdt_reint_striped_unlock(info, obj, lh, einfo, decref);
1675         }
1676 }
1677
1678 /*
1679  * lock parents of links, and also check whether total locks don't exceed
1680  * RS_MAX_LOCKS.
1681  *
1682  * \retval      0 on success, and locks can be saved in ptlrpc_reply_stat
1683  * \retval      1 on success, but total lock count may exceed RS_MAX_LOCKS
1684  * \retval      -ev negative errno upon error
1685  */
1686 static int mdt_link_parents_lock(struct mdt_thread_info *info,
1687                                  struct mdt_object *pobj,
1688                                  const struct md_attr *ma,
1689                                  struct mdt_object *obj,
1690                                  struct mdt_lock_handle *lhp,
1691                                  struct ldlm_enqueue_info *peinfo,
1692                                  struct list_head *parent_slave_locks,
1693                                  struct list_head *link_locks)
1694 {
1695         struct mdt_device *mdt = info->mti_mdt;
1696         struct lu_buf *buf = &info->mti_big_buf;
1697         struct lu_name *lname = &info->mti_name;
1698         struct linkea_data ldata = { NULL };
1699         bool blocked = false;
1700         int local_lnkp_cnt = 0;
1701         int rc;
1702
1703         ENTRY;
1704         if (S_ISDIR(lu_object_attr(&obj->mot_obj)))
1705                 RETURN(0);
1706
1707         buf = lu_buf_check_and_alloc(buf, MAX_LINKEA_SIZE);
1708         if (buf->lb_buf == NULL)
1709                 RETURN(-ENOMEM);
1710
1711         ldata.ld_buf = buf;
1712         rc = mdt_links_read(info, obj, &ldata);
1713         if (rc) {
1714                 if (rc == -ENOENT || rc == -ENODATA)
1715                         rc = 0;
1716                 RETURN(rc);
1717         }
1718
1719         for (linkea_first_entry(&ldata); ldata.ld_lee && !rc;
1720              linkea_next_entry(&ldata)) {
1721                 struct mdt_object *lnkp;
1722                 struct mdt_sub_lock *msl;
1723                 struct lu_fid fid;
1724                 __u64 ibits;
1725
1726                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, lname,
1727                                     &fid);
1728
1729                 /* check if it's also linked to parent */
1730                 if (lu_fid_eq(mdt_object_fid(pobj), &fid)) {
1731                         CDEBUG(D_INFO, "skip parent "DFID", reovke "DNAME"\n",
1732                                PFID(&fid), PNAME(lname));
1733                         /* in case link is remote object, revoke LOOKUP lock */
1734                         rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
1735                         continue;
1736                 }
1737
1738                 lnkp = NULL;
1739
1740                 /* check if it's linked to a stripe of parent */
1741                 if (ma->ma_valid & MA_LMV) {
1742                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1743                         struct lu_fid *stripe_fid = &info->mti_tmp_fid1;
1744                         int j = 0;
1745
1746                         for (; j < le32_to_cpu(lmv->lmv_stripe_count); j++) {
1747                                 fid_le_to_cpu(stripe_fid,
1748                                               &lmv->lmv_stripe_fids[j]);
1749                                 if (lu_fid_eq(stripe_fid, &fid)) {
1750                                         CDEBUG(D_INFO, "skip stripe "DFID
1751                                                ", reovke "DNAME"\n",
1752                                                PFID(&fid), PNAME(lname));
1753                                         lnkp = mdt_object_find(info->mti_env,
1754                                                                mdt, &fid);
1755                                         if (IS_ERR(lnkp))
1756                                                 GOTO(out, rc = PTR_ERR(lnkp));
1757                                         break;
1758                                 }
1759                         }
1760
1761                         if (lnkp) {
1762                                 rc = mdt_revoke_remote_lookup_lock(info, lnkp,
1763                                                                    obj);
1764                                 mdt_object_put(info->mti_env, lnkp);
1765                                 continue;
1766                         }
1767                 }
1768
1769                 /* Check if it's already locked */
1770                 list_for_each_entry(msl, link_locks, msl_linkage) {
1771                         if (lu_fid_eq(mdt_object_fid(msl->msl_obj), &fid)) {
1772                                 CDEBUG(D_INFO,
1773                                        DFID" was locked, revoke "DNAME"\n",
1774                                        PFID(&fid), PNAME(lname));
1775                                 lnkp = msl->msl_obj;
1776                                 break;
1777                         }
1778                 }
1779
1780                 if (lnkp) {
1781                         rc = mdt_revoke_remote_lookup_lock(info, lnkp, obj);
1782                         continue;
1783                 }
1784
1785                 CDEBUG(D_INFO, "lock "DFID":"DNAME"\n",
1786                        PFID(&fid), PNAME(lname));
1787
1788                 lnkp = mdt_object_find(info->mti_env, mdt, &fid);
1789                 if (IS_ERR(lnkp)) {
1790                         CWARN("%s: cannot find obj "DFID": %ld\n",
1791                               mdt_obd_name(mdt), PFID(&fid), PTR_ERR(lnkp));
1792                         continue;
1793                 }
1794
1795                 if (!mdt_object_exists(lnkp)) {
1796                         CDEBUG(D_INFO, DFID" doesn't exist, skip "DNAME"\n",
1797                               PFID(&fid), PNAME(lname));
1798                         mdt_object_put(info->mti_env, lnkp);
1799                         continue;
1800                 }
1801
1802                 if (!mdt_object_remote(lnkp))
1803                         local_lnkp_cnt++;
1804
1805                 OBD_ALLOC_PTR(msl);
1806                 if (msl == NULL)
1807                         GOTO(out, rc = -ENOMEM);
1808
1809                 /*
1810                  * we can't follow parent-child lock order like other MD
1811                  * operations, use lock_try here to avoid deadlock, if the lock
1812                  * cannot be taken, drop all locks taken, revoke the blocked
1813                  * one, and continue processing the remaining entries, and in
1814                  * the end of the loop restart from beginning.
1815                  */
1816                 mdt_lock_pdo_init(&msl->msl_lh, LCK_PW, lname);
1817                 ibits = 0;
1818                 rc = mdt_object_lock_try(info, lnkp, &msl->msl_lh, &ibits,
1819                                          MDS_INODELOCK_UPDATE, true);
1820                 if (!(ibits & MDS_INODELOCK_UPDATE)) {
1821
1822                         CDEBUG(D_INFO, "busy lock on "DFID" "DNAME"\n",
1823                                PFID(&fid), PNAME(lname));
1824
1825                         mdt_unlock_list(info, link_locks, 1);
1826                         /* also unlock parent locks to avoid deadlock */
1827                         if (!blocked)
1828                                 mdt_migrate_object_unlock(info, pobj, lhp,
1829                                                           peinfo,
1830                                                           parent_slave_locks,
1831                                                           1);
1832
1833                         blocked = true;
1834
1835                         mdt_lock_pdo_init(&msl->msl_lh, LCK_PW, lname);
1836                         rc = mdt_object_lock(info, lnkp, &msl->msl_lh,
1837                                              MDS_INODELOCK_UPDATE);
1838                         if (rc) {
1839                                 mdt_object_put(info->mti_env, lnkp);
1840                                 OBD_FREE_PTR(msl);
1841                                 GOTO(out, rc);
1842                         }
1843
1844                         if (mdt_object_remote(lnkp)) {
1845                                 struct ldlm_lock *lock;
1846
1847                                 /*
1848                                  * for remote object, set lock cb_atomic,
1849                                  * so lock can be released in blocking_ast()
1850                                  * immediately, then the next lock_try will
1851                                  * have better chance of success.
1852                                  */
1853                                 lock = ldlm_handle2lock(
1854                                                 &msl->msl_lh.mlh_rreg_lh);
1855                                 LASSERT(lock != NULL);
1856                                 lock_res_and_lock(lock);
1857                                 ldlm_set_atomic_cb(lock);
1858                                 unlock_res_and_lock(lock);
1859                                 LDLM_LOCK_PUT(lock);
1860                         }
1861
1862                         mdt_object_unlock_put(info, lnkp, &msl->msl_lh, 1);
1863                         OBD_FREE_PTR(msl);
1864                         continue;
1865                 }
1866
1867                 INIT_LIST_HEAD(&msl->msl_linkage);
1868                 msl->msl_obj = lnkp;
1869                 list_add_tail(&msl->msl_linkage, link_locks);
1870
1871                 rc = mdt_revoke_remote_lookup_lock(info, lnkp, obj);
1872         }
1873
1874         if (blocked)
1875                 GOTO(out, rc = -EBUSY);
1876
1877         EXIT;
1878 out:
1879         if (rc) {
1880                 mdt_unlock_list(info, link_locks, rc);
1881         } else if (local_lnkp_cnt > RS_MAX_LOCKS - 5) {
1882                 CDEBUG(D_INFO, "Too many links (%d), sync operations\n",
1883                        local_lnkp_cnt);
1884                 /*
1885                  * parent may have 3 local objects: master object and 2 stripes
1886                  * (if it's being migrated too); source may have 1 local objects
1887                  * as regular file; target has 1 local object.
1888                  * Note, source may have 2 local locks if it is directory but it
1889                  * can't have hardlinks, so it is not considered here.
1890                  */
1891                 rc = 1;
1892         }
1893         return rc;
1894 }
1895
1896 static int mdt_lock_remote_slaves(struct mdt_thread_info *info,
1897                                   struct mdt_object *obj,
1898                                   const struct md_attr *ma,
1899                                   struct list_head *slave_locks)
1900 {
1901         struct mdt_device *mdt = info->mti_mdt;
1902         const struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1903         struct lu_fid *fid = &info->mti_tmp_fid1;
1904         struct mdt_object *slave;
1905         struct mdt_sub_lock *msl;
1906         int i;
1907         int rc;
1908
1909         ENTRY;
1910         LASSERT(mdt_object_remote(obj));
1911         LASSERT(ma->ma_valid & MA_LMV);
1912         LASSERT(lmv);
1913
1914         if (!lmv_is_sane(lmv))
1915                 RETURN(-EINVAL);
1916
1917         for (i = 0; i < le32_to_cpu(lmv->lmv_stripe_count); i++) {
1918                 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[i]);
1919
1920                 if (!fid_is_sane(fid))
1921                         continue;
1922
1923                 slave = mdt_object_find(info->mti_env, mdt, fid);
1924                 if (IS_ERR(slave))
1925                         GOTO(out, rc = PTR_ERR(slave));
1926
1927                 OBD_ALLOC_PTR(msl);
1928                 if (!msl) {
1929                         mdt_object_put(info->mti_env, slave);
1930                         GOTO(out, rc = -ENOMEM);
1931                 }
1932
1933                 mdt_lock_reg_init(&msl->msl_lh, LCK_EX);
1934                 rc = mdt_reint_object_lock(info, slave, &msl->msl_lh,
1935                                            MDS_INODELOCK_UPDATE, true);
1936                 if (rc) {
1937                         OBD_FREE_PTR(msl);
1938                         mdt_object_put(info->mti_env, slave);
1939                         GOTO(out, rc);
1940                 }
1941
1942                 INIT_LIST_HEAD(&msl->msl_linkage);
1943                 msl->msl_obj = slave;
1944                 list_add_tail(&msl->msl_linkage, slave_locks);
1945         }
1946         EXIT;
1947
1948 out:
1949         if (rc)
1950                 mdt_unlock_list(info, slave_locks, rc);
1951         return rc;
1952 }
1953
1954 /* lock parent and its stripes */
1955 static int mdt_migrate_parent_lock(struct mdt_thread_info *info,
1956                                    struct mdt_object *obj,
1957                                    const struct md_attr *ma,
1958                                    struct mdt_lock_handle *lh,
1959                                    struct ldlm_enqueue_info *einfo,
1960                                    struct list_head *slave_locks)
1961 {
1962         int rc;
1963
1964         if (mdt_object_remote(obj)) {
1965                 rc = mdt_remote_object_lock(info, obj, mdt_object_fid(obj),
1966                                             &lh->mlh_rreg_lh, LCK_PW,
1967                                             MDS_INODELOCK_UPDATE, false);
1968                 if (rc != ELDLM_OK)
1969                         return rc;
1970
1971                 /*
1972                  * if obj is remote and striped, lock its stripes explicitly
1973                  * because it's not striped in LOD layer on this MDT.
1974                  */
1975                 if (ma->ma_valid & MA_LMV) {
1976                         rc = mdt_lock_remote_slaves(info, obj, ma, slave_locks);
1977                         if (rc)
1978                                 mdt_object_unlock(info, obj, lh, rc);
1979                 }
1980         } else {
1981                 rc = mdt_reint_striped_lock(info, obj, lh, MDS_INODELOCK_UPDATE,
1982                                             einfo, true);
1983         }
1984
1985         return rc;
1986 }
1987
1988 /*
1989  * in migration, object may be remote, and we need take full lock of it and its
1990  * stripes if it's directory, besides, object may be a remote object on its
1991  * parent, revoke its LOOKUP lock on where its parent is located.
1992  */
1993 static int mdt_migrate_object_lock(struct mdt_thread_info *info,
1994                                    struct mdt_object *pobj,
1995                                    struct mdt_object *obj,
1996                                    struct mdt_lock_handle *lh,
1997                                    struct ldlm_enqueue_info *einfo,
1998                                    struct list_head *slave_locks)
1999 {
2000         int rc;
2001
2002         if (mdt_object_remote(obj)) {
2003                 rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
2004                 if (rc)
2005                         return rc;
2006
2007                 rc = mdt_remote_object_lock(info, obj, mdt_object_fid(obj),
2008                                             &lh->mlh_rreg_lh, LCK_EX,
2009                                             MDS_INODELOCK_FULL, false);
2010                 if (rc != ELDLM_OK)
2011                         return rc;
2012
2013                 /*
2014                  * if obj is remote and striped, lock its stripes explicitly
2015                  * because it's not striped in LOD layer on this MDT.
2016                  */
2017                 if (S_ISDIR(lu_object_attr(&obj->mot_obj))) {
2018                         struct md_attr *ma = &info->mti_attr;
2019
2020                         rc = mdt_stripe_get(info, obj, ma, XATTR_NAME_LMV);
2021                         if (rc) {
2022                                 mdt_object_unlock(info, obj, lh, rc);
2023                                 return rc;
2024                         }
2025
2026                         if (ma->ma_valid & MA_LMV) {
2027                                 rc = mdt_lock_remote_slaves(info, obj, ma,
2028                                                             slave_locks);
2029                                 if (rc)
2030                                         mdt_object_unlock(info, obj, lh, rc);
2031                         }
2032                 }
2033         } else {
2034                 if (mdt_object_remote(pobj)) {
2035                         rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
2036                         if (rc)
2037                                 return rc;
2038                 }
2039
2040                 rc = mdt_reint_striped_lock(info, obj, lh, MDS_INODELOCK_FULL,
2041                                             einfo, true);
2042         }
2043
2044         return rc;
2045 }
2046
2047 /*
2048  * lookup source by name, if parent is striped directory, we need to find the
2049  * corresponding stripe where source is located, and then lookup there.
2050  *
2051  * besides, if parent is migrating too, and file is already in target stripe,
2052  * this should be a redo of 'lfs migrate' on client side.
2053  */
2054 static int mdt_migrate_lookup(struct mdt_thread_info *info,
2055                               struct mdt_object *pobj,
2056                               const struct md_attr *ma,
2057                               const struct lu_name *lname,
2058                               struct mdt_object **spobj,
2059                               struct mdt_object **sobj)
2060 {
2061         const struct lu_env *env = info->mti_env;
2062         struct lu_fid *fid = &info->mti_tmp_fid1;
2063         struct mdt_object *stripe;
2064         int rc;
2065
2066         if (ma->ma_valid & MA_LMV) {
2067                 /* if parent is striped, lookup on corresponding stripe */
2068                 struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
2069
2070                 if (!lmv_is_sane(lmv))
2071                         return -EBADF;
2072
2073                 rc = lmv_name_to_stripe_index_old(lmv, lname->ln_name,
2074                                                   lname->ln_namelen);
2075                 if (rc < 0)
2076                         return rc;
2077
2078                 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[rc]);
2079
2080                 stripe = mdt_object_find(env, info->mti_mdt, fid);
2081                 if (IS_ERR(stripe))
2082                         return PTR_ERR(stripe);
2083
2084                 fid_zero(fid);
2085                 rc = mdo_lookup(env, mdt_object_child(stripe), lname, fid,
2086                                 &info->mti_spec);
2087                 if (rc == -ENOENT && lmv_is_layout_changing(lmv)) {
2088                         /*
2089                          * if parent layout is changeing, and lookup child
2090                          * failed on source stripe, lookup again on target
2091                          * stripe, if it exists, it means previous migration
2092                          * was interrupted, and current file was migrated
2093                          * already.
2094                          */
2095                         mdt_object_put(env, stripe);
2096
2097                         rc = lmv_name_to_stripe_index(lmv, lname->ln_name,
2098                                                       lname->ln_namelen);
2099                         if (rc < 0)
2100                                 return rc;
2101
2102                         fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[rc]);
2103
2104                         stripe = mdt_object_find(env, info->mti_mdt, fid);
2105                         if (IS_ERR(stripe))
2106                                 return PTR_ERR(stripe);
2107
2108                         fid_zero(fid);
2109                         rc = mdo_lookup(env, mdt_object_child(stripe), lname,
2110                                         fid, &info->mti_spec);
2111                         mdt_object_put(env, stripe);
2112                         return rc ?: -EALREADY;
2113                 } else if (rc) {
2114                         mdt_object_put(env, stripe);
2115                         return rc;
2116                 }
2117         } else {
2118                 fid_zero(fid);
2119                 rc = mdo_lookup(env, mdt_object_child(pobj), lname, fid,
2120                                 &info->mti_spec);
2121                 if (rc)
2122                         return rc;
2123
2124                 stripe = pobj;
2125                 mdt_object_get(env, stripe);
2126         }
2127
2128         *spobj = stripe;
2129
2130         *sobj = mdt_object_find(env, info->mti_mdt, fid);
2131         if (IS_ERR(*sobj)) {
2132                 mdt_object_put(env, stripe);
2133                 rc = PTR_ERR(*sobj);
2134                 *spobj = NULL;
2135                 *sobj = NULL;
2136         }
2137
2138         return rc;
2139 }
2140
2141 /* end lease and close file for regular file */
2142 static int mdd_migrate_close(struct mdt_thread_info *info,
2143                              struct mdt_object *obj)
2144 {
2145         struct close_data *data;
2146         struct mdt_body *repbody;
2147         struct ldlm_lock *lease;
2148         int rc;
2149         int rc2;
2150
2151         rc = -EPROTO;
2152         if (!req_capsule_field_present(info->mti_pill, &RMF_MDT_EPOCH,
2153                                       RCL_CLIENT) ||
2154             !req_capsule_field_present(info->mti_pill, &RMF_CLOSE_DATA,
2155                                       RCL_CLIENT))
2156                 goto close;
2157
2158         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2159         if (!data)
2160                 goto close;
2161
2162         rc = -ESTALE;
2163         lease = ldlm_handle2lock(&data->cd_handle);
2164         if (!lease)
2165                 goto close;
2166
2167         /* check if the lease was already canceled */
2168         lock_res_and_lock(lease);
2169         rc = ldlm_is_cancel(lease);
2170         unlock_res_and_lock(lease);
2171
2172         if (rc) {
2173                 rc = -EAGAIN;
2174                 LDLM_DEBUG(lease, DFID" lease broken",
2175                            PFID(mdt_object_fid(obj)));
2176         }
2177
2178         /*
2179          * cancel server side lease, client side counterpart should have been
2180          * cancelled, it's okay to cancel it now as we've held mot_open_sem.
2181          */
2182         ldlm_lock_cancel(lease);
2183         ldlm_reprocess_all(lease->l_resource,
2184                            lease->l_policy_data.l_inodebits.bits);
2185         LDLM_LOCK_PUT(lease);
2186
2187 close:
2188         rc2 = mdt_close_internal(info, mdt_info_req(info), NULL);
2189         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2190         repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2191
2192         return rc ?: rc2;
2193 }
2194
2195 /*
2196  * migrate file in below steps:
2197  *  1. lock parent and its stripes
2198  *  2. lookup source by name
2199  *  3. lock parents of source links if source is not directory
2200  *  4. reject if source is in HSM
2201  *  5. take source open_sem and close file if source is regular file
2202  *  6. lock source and its stripes if it's directory
2203  *  7. lock target so subsequent change to it can trigger COS
2204  *  8. migrate file
2205  *  9. unlock above locks
2206  * 10. sync device if source has links
2207  */
2208 int mdt_reint_migrate(struct mdt_thread_info *info,
2209                       struct mdt_lock_handle *unused)
2210 {
2211         const struct lu_env *env = info->mti_env;
2212         struct mdt_device *mdt = info->mti_mdt;
2213         struct ptlrpc_request *req = mdt_info_req(info);
2214         struct mdt_reint_record *rr = &info->mti_rr;
2215         struct lu_ucred *uc = mdt_ucred(info);
2216         struct md_attr *ma = &info->mti_attr;
2217         struct ldlm_enqueue_info *peinfo = &info->mti_einfo[0];
2218         struct ldlm_enqueue_info *seinfo = &info->mti_einfo[1];
2219         struct mdt_object *pobj;
2220         struct mdt_object *spobj = NULL;
2221         struct mdt_object *sobj = NULL;
2222         struct mdt_object *tobj;
2223         struct lustre_handle rename_lh = { 0 };
2224         struct mdt_lock_handle *lhp;
2225         struct mdt_lock_handle *lhs;
2226         struct mdt_lock_handle *lht;
2227         LIST_HEAD(parent_slave_locks);
2228         LIST_HEAD(child_slave_locks);
2229         LIST_HEAD(link_locks);
2230         int lock_retries = 5;
2231         bool open_sem_locked = false;
2232         bool do_sync = false;
2233         int rc;
2234
2235         ENTRY;
2236         CDEBUG(D_INODE, "migrate "DFID"/"DNAME" to "DFID"\n", PFID(rr->rr_fid1),
2237                PNAME(&rr->rr_name), PFID(rr->rr_fid2));
2238
2239         if (info->mti_dlm_req)
2240                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
2241
2242         if (!fid_is_md_operative(rr->rr_fid1) ||
2243             !fid_is_md_operative(rr->rr_fid2))
2244                 RETURN(-EPERM);
2245
2246         /* don't allow migrate . or .. */
2247         if (lu_name_is_dot_or_dotdot(&rr->rr_name))
2248                 RETURN(-EBUSY);
2249
2250         if (!mdt->mdt_enable_remote_dir || !mdt->mdt_enable_dir_migration)
2251                 RETURN(-EPERM);
2252
2253         if (uc && !cap_raised(uc->uc_cap, CAP_SYS_ADMIN) &&
2254             uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
2255             mdt->mdt_enable_remote_dir_gid != -1)
2256                 RETURN(-EPERM);
2257
2258         /*
2259          * Note: do not enqueue rename lock for replay request, because
2260          * if other MDT holds rename lock, but being blocked to wait for
2261          * this MDT to finish its recovery, and the failover MDT can not
2262          * get rename lock, which will cause deadlock.
2263          *
2264          * req is NULL if this is called by directory auto-split.
2265          */
2266         if (req && !req_is_replay(req)) {
2267                 rc = mdt_rename_lock(info, &rename_lh);
2268                 if (rc != 0) {
2269                         CERROR("%s: can't lock FS for rename: rc = %d\n",
2270                                mdt_obd_name(info->mti_mdt), rc);
2271                         RETURN(rc);
2272                 }
2273         }
2274
2275         /* pobj is master object of parent */
2276         pobj = mdt_object_find(env, mdt, rr->rr_fid1);
2277         if (IS_ERR(pobj))
2278                 GOTO(unlock_rename, rc = PTR_ERR(pobj));
2279
2280         if (req) {
2281                 rc = mdt_version_get_check(info, pobj, 0);
2282                 if (rc)
2283                         GOTO(put_parent, rc);
2284         }
2285
2286         if (!mdt_object_exists(pobj))
2287                 GOTO(put_parent, rc = -ENOENT);
2288
2289         if (!S_ISDIR(lu_object_attr(&pobj->mot_obj)))
2290                 GOTO(put_parent, rc = -ENOTDIR);
2291
2292         rc = mdt_stripe_get(info, pobj, ma, XATTR_NAME_LMV);
2293         if (rc)
2294                 GOTO(put_parent, rc);
2295
2296 lock_parent:
2297         /* lock parent object */
2298         lhp = &info->mti_lh[MDT_LH_PARENT];
2299         mdt_lock_reg_init(lhp, LCK_PW);
2300         rc = mdt_migrate_parent_lock(info, pobj, ma, lhp, peinfo,
2301                                      &parent_slave_locks);
2302         if (rc)
2303                 GOTO(put_parent, rc);
2304
2305         /*
2306          * spobj is the corresponding stripe against name if pobj is striped
2307          * directory, which is the real parent, and no need to lock, because
2308          * we've taken full lock of pobj.
2309          */
2310         rc = mdt_migrate_lookup(info, pobj, ma, &rr->rr_name, &spobj, &sobj);
2311         if (rc)
2312                 GOTO(unlock_parent, rc);
2313
2314         /* lock parents of source links, and revoke LOOKUP lock of links */
2315         rc = mdt_link_parents_lock(info, pobj, ma, sobj, lhp, peinfo,
2316                                    &parent_slave_locks, &link_locks);
2317         if (rc == -EBUSY && lock_retries-- > 0) {
2318                 mdt_object_put(env, sobj);
2319                 mdt_object_put(env, spobj);
2320                 goto lock_parent;
2321         }
2322
2323         if (rc < 0)
2324                 GOTO(put_source, rc);
2325
2326         /*
2327          * RS_MAX_LOCKS is the limit of number of locks that can be saved along
2328          * with one request, if total lock count exceeds this limit, we will
2329          * drop all locks after migration, and synchronous device in the end.
2330          */
2331         do_sync = rc;
2332
2333         /* TODO: DoM migration is not supported, migrate dirent only */
2334         if (S_ISREG(lu_object_attr(&sobj->mot_obj))) {
2335                 rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LOV);
2336                 if (rc)
2337                         GOTO(unlock_links, rc);
2338
2339                 if (ma->ma_valid & MA_LOV && mdt_lmm_dom_stripesize(ma->ma_lmm))
2340                         info->mti_spec.sp_migrate_nsonly = 1;
2341         } else if (S_ISDIR(lu_object_attr(&sobj->mot_obj))) {
2342                 rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LMV);
2343                 if (rc)
2344                         GOTO(unlock_links, rc);
2345
2346                 /* race with restripe/auto-split? */
2347                 if ((ma->ma_valid & MA_LMV) &&
2348                     lmv_is_restriping(&ma->ma_lmv->lmv_md_v1))
2349                         GOTO(unlock_links, rc = -EBUSY);
2350         }
2351
2352         /* if migration HSM is allowed */
2353         if (!mdt->mdt_opts.mo_migrate_hsm_allowed) {
2354                 ma->ma_need = MA_HSM;
2355                 ma->ma_valid = 0;
2356                 rc = mdt_attr_get_complex(info, sobj, ma);
2357                 if (rc)
2358                         GOTO(unlock_links, rc);
2359
2360                 if ((ma->ma_valid & MA_HSM) && ma->ma_hsm.mh_flags != 0)
2361                         GOTO(unlock_links, rc = -EOPNOTSUPP);
2362         }
2363
2364         /* end lease and close file for regular file */
2365         if (info->mti_spec.sp_migrate_close) {
2366                 /* try to hold open_sem so that nobody else can open the file */
2367                 if (!down_write_trylock(&sobj->mot_open_sem)) {
2368                         /* close anyway */
2369                         mdd_migrate_close(info, sobj);
2370                         GOTO(unlock_links, rc = -EBUSY);
2371                 } else {
2372                         open_sem_locked = true;
2373                         rc = mdd_migrate_close(info, sobj);
2374                         if (rc)
2375                                 GOTO(unlock_open_sem, rc);
2376                 }
2377         }
2378
2379         /* lock source */
2380         lhs = &info->mti_lh[MDT_LH_OLD];
2381         mdt_lock_reg_init(lhs, LCK_EX);
2382         rc = mdt_migrate_object_lock(info, spobj, sobj, lhs, seinfo,
2383                                      &child_slave_locks);
2384         if (rc)
2385                 GOTO(unlock_open_sem, rc);
2386
2387         /* lock target */
2388         tobj = mdt_object_find(env, mdt, rr->rr_fid2);
2389         if (IS_ERR(tobj))
2390                 GOTO(unlock_source, rc = PTR_ERR(tobj));
2391
2392         lht = &info->mti_lh[MDT_LH_NEW];
2393         mdt_lock_reg_init(lht, LCK_EX);
2394         rc = mdt_reint_object_lock(info, tobj, lht, MDS_INODELOCK_FULL, true);
2395         if (rc)
2396                 GOTO(put_target, rc);
2397
2398         /* Don't do lookup sanity check. We know name doesn't exist. */
2399         info->mti_spec.sp_cr_lookup = 0;
2400         info->mti_spec.sp_feat = &dt_directory_features;
2401
2402         rc = mdo_migrate(env, mdt_object_child(pobj),
2403                          mdt_object_child(sobj), &rr->rr_name,
2404                          mdt_object_child(tobj),
2405                          &info->mti_spec, ma);
2406         if (!rc)
2407                 lprocfs_counter_incr(mdt->mdt_lu_dev.ld_obd->obd_md_stats,
2408                                      LPROC_MDT_MIGRATE + LPROC_MD_LAST_OPC);
2409         EXIT;
2410
2411         mdt_object_unlock(info, tobj, lht, rc);
2412 put_target:
2413         mdt_object_put(env, tobj);
2414 unlock_source:
2415         mdt_migrate_object_unlock(info, sobj, lhs, seinfo,
2416                                   &child_slave_locks, rc);
2417 unlock_open_sem:
2418         if (open_sem_locked)
2419                 up_write(&sobj->mot_open_sem);
2420 unlock_links:
2421         /* if we've got too many locks to save into RPC,
2422          * then just commit before the locks are released
2423          */
2424         if (!rc && do_sync)
2425                 mdt_device_sync(env, mdt);
2426         mdt_unlock_list(info, &link_locks, do_sync ? 1 : rc);
2427 put_source:
2428         mdt_object_put(env, sobj);
2429         mdt_object_put(env, spobj);
2430 unlock_parent:
2431         mdt_migrate_object_unlock(info, pobj, lhp, peinfo,
2432                                   &parent_slave_locks, rc);
2433 put_parent:
2434         mdt_object_put(env, pobj);
2435 unlock_rename:
2436         if (lustre_handle_is_used(&rename_lh))
2437                 mdt_rename_unlock(&rename_lh);
2438
2439         return rc;
2440 }
2441
2442 static int mdt_object_lock_save(struct mdt_thread_info *info,
2443                                 struct mdt_object *dir,
2444                                 struct mdt_lock_handle *lh,
2445                                 int idx, bool cos_incompat)
2446 {
2447         int rc;
2448
2449         /* we lock the target dir if it is local */
2450         rc = mdt_reint_object_lock(info, dir, lh, MDS_INODELOCK_UPDATE,
2451                                    cos_incompat);
2452         if (rc != 0)
2453                 return rc;
2454
2455         /* get and save correct version after locking */
2456         mdt_version_get_save(info, dir, idx);
2457         return 0;
2458 }
2459
2460 /*
2461  * determine lock order of sobj and tobj
2462  *
2463  * there are two situations we need to lock tobj before sobj:
2464  * 1. sobj is child of tobj
2465  * 2. sobj and tobj are stripes of a directory, and stripe index of sobj is
2466  *    larger than that of tobj
2467  *
2468  * \retval      1 lock tobj before sobj
2469  * \retval      0 lock sobj before tobj
2470  * \retval      -ev negative errno upon error
2471  */
2472 static int mdt_rename_determine_lock_order(struct mdt_thread_info *info,
2473                                            struct mdt_object *sobj,
2474                                            struct mdt_object *tobj)
2475 {
2476         struct md_attr *ma = &info->mti_attr;
2477         struct lu_fid *spfid = &info->mti_tmp_fid1;
2478         struct lu_fid *tpfid = &info->mti_tmp_fid2;
2479         struct lmv_mds_md_v1 *lmv;
2480         __u32 sindex;
2481         __u32 tindex;
2482         int rc;
2483
2484         /* sobj and tobj are the same */
2485         if (sobj == tobj)
2486                 return 0;
2487
2488         if (fid_is_root(mdt_object_fid(sobj)))
2489                 return 0;
2490
2491         if (fid_is_root(mdt_object_fid(tobj)))
2492                 return 1;
2493
2494         /* check whether sobj is child of tobj */
2495         rc = mdo_is_subdir(info->mti_env, mdt_object_child(sobj),
2496                            mdt_object_fid(tobj));
2497         if (rc < 0)
2498                 return rc;
2499
2500         if (rc == 1)
2501                 return 1;
2502
2503         /* check whether sobj and tobj are children of the same parent */
2504         rc = mdt_attr_get_pfid(info, sobj, spfid);
2505         if (rc)
2506                 return rc;
2507
2508         rc = mdt_attr_get_pfid(info, tobj, tpfid);
2509         if (rc)
2510                 return rc;
2511
2512         if (!lu_fid_eq(spfid, tpfid))
2513                 return 0;
2514
2515         /* check whether sobj and tobj are sibling stripes */
2516         rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LMV);
2517         if (rc)
2518                 return rc;
2519
2520         if (!(ma->ma_valid & MA_LMV))
2521                 return 0;
2522
2523         lmv = &ma->ma_lmv->lmv_md_v1;
2524         if (!(le32_to_cpu(lmv->lmv_magic) & LMV_MAGIC_STRIPE))
2525                 return 0;
2526         sindex = le32_to_cpu(lmv->lmv_master_mdt_index);
2527
2528         ma->ma_valid = 0;
2529         rc = mdt_stripe_get(info, tobj, ma, XATTR_NAME_LMV);
2530         if (rc)
2531                 return rc;
2532
2533         if (!(ma->ma_valid & MA_LMV))
2534                 return -ENODATA;
2535
2536         lmv = &ma->ma_lmv->lmv_md_v1;
2537         if (!(le32_to_cpu(lmv->lmv_magic) & LMV_MAGIC_STRIPE))
2538                 return -EINVAL;
2539         tindex = le32_to_cpu(lmv->lmv_master_mdt_index);
2540
2541         /* check stripe index of sobj and tobj */
2542         if (sindex == tindex)
2543                 return -EINVAL;
2544
2545         return sindex < tindex ? 0 : 1;
2546 }
2547
2548 /*
2549  * lock rename source object.
2550  *
2551  * Both source and source parent may be remote, and source may be a remote
2552  * object on source parent, to avoid overriding lock handle, store remote
2553  * LOOKUP lock separately in @lhr.
2554  *
2555  * \retval      0 on success
2556  * \retval      -ev negative errno upon error
2557  */
2558 static int mdt_rename_source_lock(struct mdt_thread_info *info,
2559                                   struct mdt_object *parent,
2560                                   struct mdt_object *child,
2561                                   struct mdt_lock_handle *lhc,
2562                                   struct mdt_lock_handle *lhr,
2563                                   __u64 ibits,
2564                                   bool cos_incompat)
2565 {
2566         int rc;
2567
2568         rc = mdt_is_remote_object(info, parent, child);
2569         if (rc < 0)
2570                 return rc;
2571
2572         if (rc) {
2573                 /* enqueue remote LOOKUP lock from the parent MDT */
2574                 __u64 rmt_ibits = MDS_INODELOCK_LOOKUP;
2575
2576                 if (mdt_object_remote(parent)) {
2577                         rc = mdt_remote_object_lock(info, parent,
2578                                                     mdt_object_fid(child),
2579                                                     &lhr->mlh_rreg_lh,
2580                                                     lhr->mlh_rreg_mode,
2581                                                     rmt_ibits, false);
2582                         if (rc != ELDLM_OK)
2583                                 return rc;
2584                 } else {
2585                         LASSERT(mdt_object_remote(child));
2586                         rc = mdt_object_local_lock(info, child, lhr,
2587                                                    &rmt_ibits, 0, true);
2588                         if (rc < 0)
2589                                 return rc;
2590                 }
2591
2592                 ibits &= ~MDS_INODELOCK_LOOKUP;
2593         }
2594
2595         if (mdt_object_remote(child)) {
2596                 rc = mdt_remote_object_lock(info, child, mdt_object_fid(child),
2597                                             &lhc->mlh_rreg_lh,
2598                                             lhc->mlh_rreg_mode,
2599                                             ibits, false);
2600                 if (rc == ELDLM_OK)
2601                         rc = 0;
2602         } else {
2603                 rc = mdt_reint_object_lock(info, child, lhc, ibits,
2604                                            cos_incompat);
2605         }
2606
2607         if (!rc)
2608                 mdt_object_unlock(info, child, lhr, rc);
2609
2610         return rc;
2611 }
2612
2613 /*
2614  * VBR: rename versions in reply: 0 - srcdir parent; 1 - tgtdir parent;
2615  * 2 - srcdir child; 3 - tgtdir child.
2616  * Update on disk version of srcdir child.
2617  */
2618 static int mdt_reint_rename(struct mdt_thread_info *info,
2619                             struct mdt_lock_handle *unused)
2620 {
2621         struct mdt_device *mdt = info->mti_mdt;
2622         struct mdt_reint_record *rr = &info->mti_rr;
2623         struct md_attr *ma = &info->mti_attr;
2624         struct ptlrpc_request *req = mdt_info_req(info);
2625         struct mdt_object *msrcdir = NULL;
2626         struct mdt_object *mtgtdir = NULL;
2627         struct mdt_object *mold;
2628         struct mdt_object *mnew = NULL;
2629         struct lustre_handle rename_lh = { 0 };
2630         struct mdt_lock_handle *lh_srcdirp;
2631         struct mdt_lock_handle *lh_tgtdirp;
2632         struct mdt_lock_handle *lh_oldp = NULL;
2633         struct mdt_lock_handle *lh_rmt = NULL;
2634         struct mdt_lock_handle *lh_newp = NULL;
2635         struct lu_fid *old_fid = &info->mti_tmp_fid1;
2636         struct lu_fid *new_fid = &info->mti_tmp_fid2;
2637         __u64 lock_ibits;
2638         bool reverse = false, discard = false;
2639         bool cos_incompat;
2640         ktime_t kstart = ktime_get();
2641         int rc;
2642
2643         ENTRY;
2644         DEBUG_REQ(D_INODE, req, "rename "DFID"/"DNAME" to "DFID"/"DNAME,
2645                   PFID(rr->rr_fid1), PNAME(&rr->rr_name),
2646                   PFID(rr->rr_fid2), PNAME(&rr->rr_tgt_name));
2647
2648         if (info->mti_dlm_req)
2649                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
2650
2651         if (!fid_is_md_operative(rr->rr_fid1) ||
2652             !fid_is_md_operative(rr->rr_fid2))
2653                 RETURN(-EPERM);
2654
2655         /* find both parents. */
2656         msrcdir = mdt_parent_find_check(info, rr->rr_fid1, 0);
2657         if (IS_ERR(msrcdir))
2658                 RETURN(PTR_ERR(msrcdir));
2659
2660         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
2661
2662         if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
2663                 mtgtdir = msrcdir;
2664                 mdt_object_get(info->mti_env, mtgtdir);
2665         } else {
2666                 mtgtdir = mdt_parent_find_check(info, rr->rr_fid2, 1);
2667                 if (IS_ERR(mtgtdir))
2668                         GOTO(out_put_srcdir, rc = PTR_ERR(mtgtdir));
2669         }
2670
2671         /*
2672          * Note: do not enqueue rename lock for replay request, because
2673          * if other MDT holds rename lock, but being blocked to wait for
2674          * this MDT to finish its recovery, and the failover MDT can not
2675          * get rename lock, which will cause deadlock.
2676          */
2677         if (!req_is_replay(req)) {
2678                 /*
2679                  * Normally rename RPC is handled on the MDT with the target
2680                  * directory (if target exists, it's on the MDT with the
2681                  * target), if the source directory is remote, it's a hint that
2682                  * source is remote too (this may not be true, but it won't
2683                  * cause any issue), return -EXDEV early to avoid taking
2684                  * rename_lock.
2685                  */
2686                 if (!mdt->mdt_enable_remote_rename &&
2687                     mdt_object_remote(msrcdir))
2688                         GOTO(out_put_tgtdir, rc = -EXDEV);
2689
2690                 /* This might be further relaxed in the future for regular file
2691                  * renames in different source and target parents. Start with
2692                  * only same-directory renames for simplicity and because this
2693                  * is by far the most the common use case.
2694                  */
2695                 if (msrcdir != mtgtdir) {
2696                         rc = mdt_rename_lock(info, &rename_lh);
2697                         if (rc != 0) {
2698                                 CERROR("%s: cannot lock for rename: rc = %d\n",
2699                                        mdt_obd_name(mdt), rc);
2700                                 GOTO(out_put_tgtdir, rc);
2701                         }
2702                 } else {
2703                         CDEBUG(D_INFO, "%s: samedir rename "DFID"/"DNAME"\n",
2704                                mdt_obd_name(mdt), PFID(rr->rr_fid1),
2705                                PNAME(&rr->rr_name));
2706                 }
2707         }
2708
2709         rc = mdt_rename_determine_lock_order(info, msrcdir, mtgtdir);
2710         if (rc < 0)
2711                 GOTO(out_unlock_rename, rc);
2712
2713         reverse = rc;
2714
2715         /* source needs to be looked up after locking source parent, otherwise
2716          * this rename may race with unlink source, and cause rename hang, see
2717          * sanityn.sh 55b, so check parents first, if later we found source is
2718          * remote, relock parents.
2719          */
2720         cos_incompat = (mdt_object_remote(msrcdir) ||
2721                         mdt_object_remote(mtgtdir));
2722
2723         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
2724
2725         /* lock parents in the proper order. */
2726         lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
2727         lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
2728
2729         OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
2730         OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
2731 relock:
2732         mdt_lock_pdo_init(lh_srcdirp, LCK_PW, &rr->rr_name);
2733         mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, &rr->rr_tgt_name);
2734
2735         if (reverse) {
2736                 rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
2737                                           cos_incompat);
2738                 if (rc)
2739                         GOTO(out_unlock_rename, rc);
2740
2741                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
2742
2743                 rc = mdt_object_lock_save(info, msrcdir, lh_srcdirp, 0,
2744                                           cos_incompat);
2745                 if (rc != 0) {
2746                         mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
2747                         GOTO(out_unlock_rename, rc);
2748                 }
2749         } else {
2750                 rc = mdt_object_lock_save(info, msrcdir, lh_srcdirp, 0,
2751                                           cos_incompat);
2752                 if (rc)
2753                         GOTO(out_unlock_rename, rc);
2754
2755                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
2756
2757                 if (mtgtdir != msrcdir) {
2758                         rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
2759                                                   cos_incompat);
2760                 } else if (!mdt_object_remote(mtgtdir) &&
2761                            lh_srcdirp->mlh_pdo_hash !=
2762                            lh_tgtdirp->mlh_pdo_hash) {
2763                         rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
2764                                                 MDS_INODELOCK_UPDATE,
2765                                                 cos_incompat);
2766                         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
2767                 }
2768                 if (rc != 0) {
2769                         mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
2770                         GOTO(out_unlock_rename, rc);
2771                 }
2772         }
2773
2774         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
2775         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME2, 5);
2776
2777         /* find mold object. */
2778         fid_zero(old_fid);
2779         rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
2780         if (rc != 0)
2781                 GOTO(out_unlock_parents, rc);
2782
2783         if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
2784                 GOTO(out_unlock_parents, rc = -EINVAL);
2785
2786         if (!fid_is_md_operative(old_fid))
2787                 GOTO(out_unlock_parents, rc = -EPERM);
2788
2789         mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
2790         if (IS_ERR(mold))
2791                 GOTO(out_unlock_parents, rc = PTR_ERR(mold));
2792
2793         if (!mdt_object_exists(mold)) {
2794                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2795                                 &mold->mot_obj,
2796                                 "object does not exist");
2797                 GOTO(out_put_old, rc = -ENOENT);
2798         }
2799
2800         if (mdt_object_remote(mold) && !mdt->mdt_enable_remote_rename)
2801                 GOTO(out_put_old, rc = -EXDEV);
2802
2803         /* Check if @mtgtdir is subdir of @mold, before locking child
2804          * to avoid reverse locking.
2805          */
2806         if (mtgtdir != msrcdir) {
2807                 rc = mdo_is_subdir(info->mti_env, mdt_object_child(mtgtdir),
2808                                    old_fid);
2809                 if (rc) {
2810                         if (rc == 1)
2811                                 rc = -EINVAL;
2812                         GOTO(out_put_old, rc);
2813                 }
2814         }
2815
2816         tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mold));
2817         /* save version after locking */
2818         mdt_version_get_save(info, mold, 2);
2819
2820         if (!cos_incompat && mdt_object_remote(mold)) {
2821                 cos_incompat = true;
2822                 mdt_object_put(info->mti_env, mold);
2823                 mdt_object_unlock(info, mtgtdir, lh_tgtdirp, -EAGAIN);
2824                 mdt_object_unlock(info, msrcdir, lh_srcdirp, -EAGAIN);
2825                 goto relock;
2826         }
2827
2828         /* find mnew object:
2829          * mnew target object may not exist now
2830          * lookup with version checking
2831          */
2832         fid_zero(new_fid);
2833         rc = mdt_lookup_version_check(info, mtgtdir, &rr->rr_tgt_name, new_fid,
2834                                       3);
2835         if (rc == 0) {
2836                 /* the new_fid should have been filled at this moment */
2837                 if (lu_fid_eq(old_fid, new_fid))
2838                         GOTO(out_put_old, rc);
2839
2840                 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
2841                     lu_fid_eq(new_fid, rr->rr_fid2))
2842                         GOTO(out_put_old, rc = -EINVAL);
2843
2844                 if (!fid_is_md_operative(new_fid))
2845                         GOTO(out_put_old, rc = -EPERM);
2846
2847                 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
2848                 if (IS_ERR(mnew))
2849                         GOTO(out_put_old, rc = PTR_ERR(mnew));
2850
2851                 if (!mdt_object_exists(mnew)) {
2852                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2853                                         &mnew->mot_obj,
2854                                         "object does not exist");
2855                         GOTO(out_put_new, rc = -ENOENT);
2856                 }
2857
2858                 if (mdt_object_remote(mnew)) {
2859                         struct mdt_body  *repbody;
2860
2861                         /* Always send rename req to the target child MDT */
2862                         repbody = req_capsule_server_get(info->mti_pill,
2863                                                          &RMF_MDT_BODY);
2864                         LASSERT(repbody != NULL);
2865                         repbody->mbo_fid1 = *new_fid;
2866                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
2867                         GOTO(out_put_new, rc = -EXDEV);
2868                 }
2869                 /* Before locking the target dir, check we do not replace
2870                  * a dir with a non-dir, otherwise it may deadlock with
2871                  * link op which tries to create a link in this dir
2872                  * back to this non-dir.
2873                  */
2874                 if (S_ISDIR(lu_object_attr(&mnew->mot_obj)) &&
2875                     !S_ISDIR(lu_object_attr(&mold->mot_obj)))
2876                         GOTO(out_put_new, rc = -EISDIR);
2877
2878                 lh_oldp = &info->mti_lh[MDT_LH_OLD];
2879                 lh_rmt = &info->mti_lh[MDT_LH_RMT];
2880                 mdt_lock_reg_init(lh_oldp, LCK_EX);
2881                 mdt_lock_reg_init(lh_rmt, LCK_EX);
2882                 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_XATTR;
2883                 rc = mdt_rename_source_lock(info, msrcdir, mold, lh_oldp,
2884                                             lh_rmt, lock_ibits, cos_incompat);
2885                 if (rc < 0)
2886                         GOTO(out_put_new, rc);
2887
2888                 /* Check if @msrcdir is subdir of @mnew, before locking child
2889                  * to avoid reverse locking.
2890                  */
2891                 if (mtgtdir != msrcdir) {
2892                         rc = mdo_is_subdir(info->mti_env,
2893                                            mdt_object_child(msrcdir), new_fid);
2894                         if (rc) {
2895                                 if (rc == 1)
2896                                         rc = -EINVAL;
2897                                 GOTO(out_unlock_old, rc);
2898                         }
2899                 }
2900
2901                 /* We used to acquire MDS_INODELOCK_FULL here but we
2902                  * can't do this now because a running HSM restore on
2903                  * the rename onto victim will hold the layout
2904                  * lock. See LU-4002.
2905                  */
2906
2907                 lh_newp = &info->mti_lh[MDT_LH_NEW];
2908                 mdt_lock_reg_init(lh_newp, LCK_EX);
2909                 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2910                 if (mdt_object_remote(mtgtdir)) {
2911                         rc = mdt_remote_object_lock(info, mtgtdir,
2912                                                     mdt_object_fid(mnew),
2913                                                     &lh_newp->mlh_rreg_lh,
2914                                                     lh_newp->mlh_rreg_mode,
2915                                                     MDS_INODELOCK_LOOKUP,
2916                                                     false);
2917                         if (rc != ELDLM_OK)
2918                                 GOTO(out_unlock_old, rc);
2919
2920                         lock_ibits &= ~MDS_INODELOCK_LOOKUP;
2921                 }
2922                 rc = mdt_reint_object_lock(info, mnew, lh_newp, lock_ibits,
2923                                            cos_incompat);
2924                 if (rc != 0)
2925                         GOTO(out_unlock_new, rc);
2926
2927                 /* get and save version after locking */
2928                 mdt_version_get_save(info, mnew, 3);
2929         } else if (rc != -ENOENT) {
2930                 GOTO(out_put_old, rc);
2931         } else {
2932                 lh_oldp = &info->mti_lh[MDT_LH_OLD];
2933                 lh_rmt = &info->mti_lh[MDT_LH_RMT];
2934                 mdt_lock_reg_init(lh_oldp, LCK_EX);
2935                 mdt_lock_reg_init(lh_rmt, LCK_EX);
2936                 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_XATTR;
2937                 rc = mdt_rename_source_lock(info, msrcdir, mold, lh_oldp,
2938                                             lh_rmt, lock_ibits, cos_incompat);
2939                 if (rc != 0)
2940                         GOTO(out_put_old, rc);
2941
2942                 mdt_enoent_version_save(info, 3);
2943         }
2944
2945         /* step 5: rename it */
2946         mdt_reint_init_ma(info, ma);
2947
2948         mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
2949                        OBD_FAIL_MDS_REINT_RENAME_WRITE);
2950
2951         if (mnew != NULL)
2952                 mutex_lock(&mnew->mot_lov_mutex);
2953
2954         rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
2955                         mdt_object_child(mtgtdir), old_fid, &rr->rr_name,
2956                         mnew != NULL ? mdt_object_child(mnew) : NULL,
2957                         &rr->rr_tgt_name, ma);
2958
2959         if (mnew != NULL)
2960                 mutex_unlock(&mnew->mot_lov_mutex);
2961
2962         /* handle last link of tgt object */
2963         if (rc == 0) {
2964                 mdt_counter_incr(req, LPROC_MDT_RENAME,
2965                                  ktime_us_delta(ktime_get(), kstart));
2966                 if (mnew) {
2967                         mdt_handle_last_unlink(info, mnew, ma);
2968                         discard = mdt_dom_check_for_discard(info, mnew);
2969                 }
2970                 mdt_rename_counter_tally(info, info->mti_mdt, req,
2971                                          msrcdir, mtgtdir,
2972                                          ktime_us_delta(ktime_get(), kstart));
2973         }
2974
2975         EXIT;
2976 out_unlock_new:
2977         if (mnew != NULL)
2978                 mdt_object_unlock(info, mnew, lh_newp, rc);
2979 out_unlock_old:
2980         mdt_object_unlock(info, NULL, lh_rmt, rc);
2981         mdt_object_unlock(info, mold, lh_oldp, rc);
2982 out_put_new:
2983         if (mnew && !discard)
2984                 mdt_object_put(info->mti_env, mnew);
2985 out_put_old:
2986         mdt_object_put(info->mti_env, mold);
2987 out_unlock_parents:
2988         mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
2989         mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
2990 out_unlock_rename:
2991         if (lustre_handle_is_used(&rename_lh))
2992                 mdt_rename_unlock(&rename_lh);
2993 out_put_tgtdir:
2994         mdt_object_put(info->mti_env, mtgtdir);
2995 out_put_srcdir:
2996         mdt_object_put(info->mti_env, msrcdir);
2997
2998         /* The DoM discard can be done right in the place above where it is
2999          * assigned, meanwhile it is done here after rename unlock due to
3000          * compatibility with old clients, for them the discard blocks
3001          * the main thread until completion. Check LU-11359 for details.
3002          */
3003         if (discard) {
3004                 mdt_dom_discard_data(info, mnew);
3005                 mdt_object_put(info->mti_env, mnew);
3006         }
3007         OBD_RACE(OBD_FAIL_MDS_LINK_RENAME_RACE);
3008         return rc;
3009 }
3010
3011 static int mdt_reint_resync(struct mdt_thread_info *info,
3012                             struct mdt_lock_handle *lhc)
3013 {
3014         struct mdt_reint_record *rr = &info->mti_rr;
3015         struct ptlrpc_request *req = mdt_info_req(info);
3016         struct md_attr *ma = &info->mti_attr;
3017         struct mdt_object *mo;
3018         struct ldlm_lock *lease;
3019         struct mdt_body *repbody;
3020         struct md_layout_change layout = { .mlc_mirror_id = rr->rr_mirror_id };
3021         bool lease_broken;
3022         int rc, rc2;
3023
3024         ENTRY;
3025         DEBUG_REQ(D_INODE, req, DFID", FLR file resync", PFID(rr->rr_fid1));
3026
3027         if (info->mti_dlm_req)
3028                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
3029
3030         mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
3031         if (IS_ERR(mo))
3032                 GOTO(out, rc = PTR_ERR(mo));
3033
3034         if (!mdt_object_exists(mo))
3035                 GOTO(out_obj, rc = -ENOENT);
3036
3037         if (!S_ISREG(lu_object_attr(&mo->mot_obj)))
3038                 GOTO(out_obj, rc = -EINVAL);
3039
3040         if (mdt_object_remote(mo))
3041                 GOTO(out_obj, rc = -EREMOTE);
3042
3043         lease = ldlm_handle2lock(rr->rr_lease_handle);
3044         if (lease == NULL)
3045                 GOTO(out_obj, rc = -ESTALE);
3046
3047         /* It's really necessary to grab open_sem and check if the lease lock
3048          * has been lost. There would exist a concurrent writer coming in and
3049          * generating some dirty data in memory cache, the writeback would fail
3050          * after the layout version is increased by MDS_REINT_RESYNC RPC.
3051          */
3052         if (!down_write_trylock(&mo->mot_open_sem))
3053                 GOTO(out_put_lease, rc = -EBUSY);
3054
3055         lock_res_and_lock(lease);
3056         lease_broken = ldlm_is_cancel(lease);
3057         unlock_res_and_lock(lease);
3058         if (lease_broken)
3059                 GOTO(out_unlock, rc = -EBUSY);
3060
3061         /* the file has yet opened by anyone else after we took the lease. */
3062         layout.mlc_opc = MD_LAYOUT_RESYNC;
3063         lhc = &info->mti_lh[MDT_LH_LOCAL];
3064         rc = mdt_layout_change(info, mo, lhc, &layout);
3065         if (rc)
3066                 GOTO(out_unlock, rc);
3067
3068         mdt_object_unlock(info, mo, lhc, 0);
3069
3070         ma->ma_need = MA_INODE;
3071         ma->ma_valid = 0;
3072         rc = mdt_attr_get_complex(info, mo, ma);
3073         if (rc != 0)
3074                 GOTO(out_unlock, rc);
3075
3076         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3077         mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
3078
3079         EXIT;
3080 out_unlock:
3081         up_write(&mo->mot_open_sem);
3082 out_put_lease:
3083         LDLM_LOCK_PUT(lease);
3084 out_obj:
3085         mdt_object_put(info->mti_env, mo);
3086 out:
3087         mdt_client_compatibility(info);
3088         rc2 = mdt_fix_reply(info);
3089         if (rc == 0)
3090                 rc = rc2;
3091         return rc;
3092 }
3093
3094 struct mdt_reinter {
3095         int (*mr_handler)(struct mdt_thread_info *, struct mdt_lock_handle *);
3096         enum lprocfs_extra_opc mr_extra_opc;
3097 };
3098
3099 static const struct mdt_reinter mdt_reinters[] = {
3100         [REINT_SETATTR] = {
3101                 .mr_handler = &mdt_reint_setattr,
3102                 .mr_extra_opc = MDS_REINT_SETATTR,
3103         },
3104         [REINT_CREATE] = {
3105                 .mr_handler = &mdt_reint_create,
3106                 .mr_extra_opc = MDS_REINT_CREATE,
3107         },
3108         [REINT_LINK] = {
3109                 .mr_handler = &mdt_reint_link,
3110                 .mr_extra_opc = MDS_REINT_LINK,
3111         },
3112         [REINT_UNLINK] = {
3113                 .mr_handler = &mdt_reint_unlink,
3114                 .mr_extra_opc = MDS_REINT_UNLINK,
3115         },
3116         [REINT_RENAME] = {
3117                 .mr_handler = &mdt_reint_rename,
3118                 .mr_extra_opc = MDS_REINT_RENAME,
3119         },
3120         [REINT_OPEN] = {
3121                 .mr_handler = &mdt_reint_open,
3122                 .mr_extra_opc = MDS_REINT_OPEN,
3123         },
3124         [REINT_SETXATTR] = {
3125                 .mr_handler = &mdt_reint_setxattr,
3126                 .mr_extra_opc = MDS_REINT_SETXATTR,
3127         },
3128         [REINT_RMENTRY] = {
3129                 .mr_handler = &mdt_reint_unlink,
3130                 .mr_extra_opc = MDS_REINT_UNLINK,
3131         },
3132         [REINT_MIGRATE] = {
3133                 .mr_handler = &mdt_reint_migrate,
3134                 .mr_extra_opc = MDS_REINT_RENAME,
3135         },
3136         [REINT_RESYNC] = {
3137                 .mr_handler = &mdt_reint_resync,
3138                 .mr_extra_opc = MDS_REINT_RESYNC,
3139         },
3140 };
3141
3142 int mdt_reint_rec(struct mdt_thread_info *info,
3143                   struct mdt_lock_handle *lhc)
3144 {
3145         const struct mdt_reinter *mr;
3146         int rc;
3147
3148         ENTRY;
3149         if (!(info->mti_rr.rr_opcode < ARRAY_SIZE(mdt_reinters)))
3150                 RETURN(-EPROTO);
3151
3152         mr = &mdt_reinters[info->mti_rr.rr_opcode];
3153         if (mr->mr_handler == NULL)
3154                 RETURN(-EPROTO);
3155
3156         rc = (*mr->mr_handler)(info, lhc);
3157
3158         lprocfs_counter_incr(ptlrpc_req2svc(mdt_info_req(info))->srv_stats,
3159                              PTLRPC_LAST_CNTR + mr->mr_extra_opc);
3160
3161         RETURN(rc);
3162 }